Determinants of modern family planning use among married adolescents in Bangladesh: Evidence from a machine learning approach

Determinants of modern family planning use among married adolescents in Bangladesh: Evidence from a machine learning approach | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Determinants of modern family planning use among married adolescents in Bangladesh: Evidence from a machine learning approach Shawkatul Islam, Md Eyah Eya, Muhammad Khairul Alam This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8806294/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 13 You are reading this latest preprint version Abstract Background: Modern family planning (FP) is essential for improving the reproductive health of adolescent women, especially in Bangladesh, where early marriage is still prevalent. Despite overall progress in family planning, modern FP use remains uneven and relatively low among married adolescent women due to social, relational, and contextual barriers. Conventional statistical approaches might not fully capture the intricate, nonlinear determinants of FP use. Therefore, this study applied and compared multiple machine learning (ML) models to predict modern FP use and identify its key determinants among married adolescent women in Bangladesh. Methods and materials: Data were obtained from the nationally representative Bangladesh Adolescent Health and Wellbeing Survey (BAHWS) 2019–20. The analysis included a weighted sample of 3,223 ever-married adolescent females aged 15 to 19 years. A variety of machine learning classification models were employed, including Logistic Regression (LR), Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Neural Network (NN), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Adaptive Boosting (AdaBoost), and Categorical Naïve Bayes (CNB). Feature selection was performed with the Boruta algorithm, and model interpretability was examined through SHapley Additive exPlanations (SHAP). Model performance was assessed using accuracy, precision, recall, F1-score, Matthews Correlation Coefficient, Cohen's Kappa, and area under the receiver operating characteristic curve (AUROC). Result: Overall, 72.7% of married adolescent women reported using a modern FP method. Ensemble-based models outperformed conventional classifiers, especially after class balancing. XGB had the best overall predictive performance after applying SMOTE, with 76.0% accuracy, 81.5% precision, 83.9% F1 Score, 37.2% MCC, 36.9% Cohen's Kappa, 86.5% recall, and an AUROC of 72.9%, followed by RF and NN models. The most important determinants of modern FP use included spousal co-residency, having given birth, joint decision-making about FP use, administrative division, younger age (15–17), household wealth, and media exposure. Conclusion: This study reveals that machine learning models, especially XGBoost, can be useful in predicting modern family planning use and identifying key determinants among married adolescent women in Bangladesh. The findings provide a data-driven foundation for policymakers and program planners to create adolescent-focused family planning programs. The results highlight the significance of regional disparities, socioeconomic inequities, and relationship factors in influencing adolescent contraceptive behavior. Integrating ML-based evidence into family planning programs may enhance targeted interventions and help to achieve Sustainable Development Goal 3.7. Modern family planning married adolescent machine learning SHAP analysis Bangladesh Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Modern family planning methods encompass a variety of contraceptive methods, including the use of oral pills, injectables, intrauterine devices (IUDs), implants, condoms, emergency contraception, and permanent forms of contraception, such as female and male sterilization [ 1 ]. The widespread use and accessibility of these methods are crucial for preventing unwanted pregnancies, reducing maternal, perinatal, and neonatal mortality, and facilitating effective birth spacing [ 2 , 3 ]. Beyond their direct health benefits, modern contraceptive methods contribute to broader social gains by giving women enough time to recover between pregnancies, improving child health and educational outcomes, and contributing to overall socioeconomic growth [ 4 ]. Among the 1.9 billion women of reproductive age (15–49 years) worldwide in 2021, an estimated 1.1 billion had a need for family planning; of these, 874 million were using modern contraceptive methods, while 164 million had an unmet need for contraception [ 5 ]. Projections indicate that by 2030, approximately 1.19 billion women will require modern family planning methods. However, the use of modern contraception varies across countries and regions [ 3 ]. The use of modern contraception among married women of reproductive age is considerably lower in developing nations (45%) than in developed countries (70%), with notable regional variation, including 56% in Latin America, 74% in East Asia, 34% in South Asia, and 14% in Africa [ 6 ]. The particularly low prevalence of modern contraceptive use in South Asia is concerning, given that Bangladesh is part of this region and continues to experience high levels of early marriage. On a global and regional scale, women under the age of 25 consistently demonstrate the lowest levels of contraceptive use and pregnancy avoidance in comparison to women aged 25–44 years. This highlights a huge unmet need among adolescents and young women [ 7 ]. Bangladesh has a large and young population, with nearly half of its 160.4 million people below the age of 25, and it continues to report one of the highest rates of child marriage globally [ 8 – 10 ]. As an increasing number of adolescents enter marriage each year, early marriage exposes young women to frequent and often unprotected sexual intercourse, thereby increasing the risk of early, rapid, and unintended pregnancies with adverse maternal and neonatal health consequences [ 11 , 12 ]. While Bangladesh has achieved significant strides in increasing the use of contraceptives on a national level, these improvements conceal underlying inequalities [ 13 ]. Married adolescent women aged 15–19 years remain a high-risk and underserved group, demonstrating much lower contraceptive use than older women due to a combination of individual, relational, and contextual factors [ 13 , 14 ]. Therefore, in order to improve reproductive health outcomes and achieve Sustainable Development Goal 3.7, which is to ensure universal access to sexual and reproductive health services, it is vital to ensure that this population has equitable access to modern family planning methods [ 15 ]. Epidemiological studies using survey data and logistic regression have established key determinants of contraceptive use in this group as parity, co-residency with a spouse, joint-decision making, exposure to the media, household income, and geographic location [ 16 – 18 ]. On the other hand, separation between spouses, particularly when that is caused by international migration of males, is a tremendous obstacle [ 17 ]. Geographically, such divisions as Mymensingh and Rajshahi are recording high usage rates as compared to Chittagong meaning that there are inequalities in terms of accessing services, social norms or program implementation [ 13 ]. Despite this progress, a critical gap remains. Traditional regression models, whilst establishing independent relationships, might not be able to reflect the non-linear, and higher-order interactions that are likely to define decision-making on contraception in reality [ 19 ]. This limitation hinders the development of predictive models that are highly accurate and necessary for the precise targeting of interventions. Machine learning (ML) offers a powerful paradigm shift. Machine learning algorithms, including ensemble methods and neural networks, proficiently model complex patterns in high-dimensional data without rigid a priori assumptions regarding variable relationships, rendering them particularly effective for predictive tasks [ 20 ]. The application of machine learning to predict reproductive health is gaining popularity worldwide. Research conducted in Ethiopia and East Africa has effectively utilized algorithms such as Random Forest and XGBoost to predict contraceptive discontinuation and non-use, indicating their capability to surpass conventional approaches [ 21 – 24 ]. In Bangladesh, initial research used machine learning to classify unintended pregnancy [ 25 ]. However, there is a significant gap in the literature. To date, no study has systematically applied and compared multiple advanced machine learning algorithms to identify and interpret the key determinants of modern family planning use among married adolescent women. By combining feature selection, explainable machine learning techniques, and comparative model evaluation, the present study addresses this gap by identifying the most influential socio-demographic, relational, and contextual determinants of modern family planning use among married adolescent women in Bangladesh. This study uses data from the 2019–20 Bangladesh Adolescent Health and Wellbeing Survey (BAHWS) to fill this gap. Specifically, the study aims to: (i) apply and systematically compare multiple machine learning models, including XGBoost, Random Forest, Neural Networks, and logistic regression, to identify the key determinants of modern family planning use among married adolescent women [ 22 , 26 – 28 ]; (ii) address class imbalance by using the Synthetic Minority Oversampling Technique (SMOTE) and assess the effect it has on model performance; and (iii) use sophisticated model interpretability techniques, especially SHapley Additive exPlanations (SHAP), to quantify and interpret the relative importance and directional effects of influential predictors [ 29 ]. The findings are intended to inform targeted public health interventions and evidence-based policymaking aligned with Sustainable Development Goal targets, with the broader objective of improving reproductive health and child survival outcomes in Bangladesh. Material and Methods Data sources The data for this study were derived from the Bangladesh Adolescent Health and Well-being Survey (BAHWS) 2019–20, a nationwide cross-sectional survey that assesses the health and well-being of adolescents aged 15–19 years in Bangladesh across a variety of factors. The BAHWS used a two-stage stratified sample strategy that included all administrative divisions and enumeration areas (EAs) throughout the country, assuring national representativeness. First, primary sampling units (PSUs) were identified, and subsequently, households were selected in a systematic manner during the second stage. PSUs were randomly classified into Type I and Type II, and questionnaires corresponding to PSU type (“Type One” or “Type Two”) were administered accordingly, following the official BAHWS sampling protocol [ 30 ]. The BAHWS collected information from ever-married adolescent women, never-married adolescent women, and never-married adolescent men aged 15–19 years. The present analysis was restricted to ever-married adolescent women, as information on modern family planning use was collected for this subgroup. Of the 72,800 households selected for the survey, 97.7% responded, and interviews were successfully completed in 67,093 households. Among eligible ever-married adolescent women aged 15–19 years, 4,926 respondents completed the interview, yielding a 97.2% response rate among 5,066 eligible individuals. For this study, we used the publicly accessible BAHWS 2019-20 dataset to extract records from ever-married adolescent women. Respondents who provided missing information on the outcome variable or key explanatory variables were excluded to ensure data quality. After these exclusions and survey weighting, the final analytic sample consisted of 3,223 ever-married teenage women aged 15 to 19, who were utilized for all descriptive analyses, machine learning model building, and regression validation. Response variable The outcome variable for this study was current use of modern family planning methods. In the BAHWS 2019–20 survey, ever-married adolescent women were asked: “Are you or your partner currently doing something or using any method to delay or avoid getting pregnant?” Respondents who answered affirmatively were further asked to specify the method currently being used. A binary outcome variable was created based on the answers to this question. Women were classified as modern family planning users (coded as 1) if they reported using any modern contraceptive method, such as female or male sterilization, intrauterine devices (IUD), injectables, implants, oral pills, male or female condoms, emergency contraceptive pills (ECP), or the lactational amenorrhea method (LAM). Women who reported not using any method or who used traditional methods, such as the rhythm (safe-period) method or withdrawal, were classified as non-users of modern family planning (coded as 0), in accordance with the World Health Organization (WHO) classification [ 31 ]. Explanatory variables Guided by prior literature [ 3 , 9 , 32 – 34 ] and data availability in the BAHWS 2019–20, the explanatory variables included age group (15–17, 18–19 years), educational attainment (0–5, 6–10, ≥ 11 grades), employment status (yes/no), involvement in family planning programs in the past three years (yes/no), current residence with husband (living together, staying elsewhere in Bangladesh, staying outside Bangladesh), decision-making authority for family planning use (joint, mainly husband, mainly respondent, others), childbirth experience (ever given birth: yes/no), household wealth quintile (lowest to highest), exposure to mass media (frequency of watching television and reading newspapers/magazines), recent exposure to family planning messages through television or newspapers, place of residence (urban/rural), and administrative division. Statistical analysis Machine learning modeling approach This study used various supervised machine learning classification models to determine the factors that influence modern family planning method utilization among ever-married adolescent women. The outcome variable was binary, indicating the use of any modern contraceptive method (yes/no). The machine learning models used were Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), AdaBoost, Neural Network (Multi-Layer Perceptron), and Categorical Naïve Bayes (CNB). The analytic dataset was randomly split into training (70%) and testing (30%) subsets using stratified sampling to maintain the distribution of modern family planning users and non-users. To ensure reproducibility, a fixed random seed (random_state = 42) was used. Data preprocessing and encoding One-hot encoding was used to transform categorical predictors in most models, whereas ordinal encoding was used in the Categorical Naïve Bayes model. Continuous variables were scaled using Min-Max normalization for models that are sensitive to feature scale, such as LR, SVM, KNN, and Neural Networks. Tree-based models were fitted with unscaled numerical features. This preprocessing was carried out using pipeline-based procedures to prevent data leaks during model training and evaluation. Handling class imbalance The outcome variable exhibited a class imbalance, with a lower proportion of modern family planning non-users (27.34% ) than users (72.66%). To address this issue, most models' training data were treated with the Synthetic Minority Over-sampling Technique (SMOTE), resulting in a balanced 50:50 class distribution. For categorical features, SMOTEN was used in combination with the Categorical Naïve Bayes model. Model performance was assessed in both SMOTE-adjusted and non-SMOTE conditions to evaluate robustness. Hyperparameter optimization We used GridSearchCV with five-fold cross-validation on the training dataset to find the best hyperparameters. Various performance metrics were evaluated during the tuning process. This method guaranteed optimal model calibration for identifying modern family planning users while addressing class imbalance. Model evaluation and validation Model performance on the held-out test dataset was assessed using accuracy, precision, recall (sensitivity), F1-score, Matthew’s correlation coefficient (MCC), Cohen's Kappa, and AUROC. Confusion matrices were created to evaluate classification errors in detail. Additional k-fold cross-validation was performed with different fold sizes (5, 10, 15, 20, 25, and 30 folds) in order to ensure robustness and generalizability. Receiver operating characteristic analysis The ROC curve graphically represents the trade-off between true positives and false positives at the various classification thresholds [ 35 ]. ROC curves were generated for all fitted models to illustrate the discrimination performance across classification thresholds. The performance of models was compared using the AUROC, with higher values indicating better discrimination between modern family planning users and non-users. Feature selection and explainability The Boruta algorithm was used to select the most important predictors of modern family planning use. Boruta compares the relevance of observed variables to randomized shadow features in a Random Forest framework, retaining only statistically significant predictors [ 36 ]. To improve model interpretability, feature importance from tree-based models and SHapley Additive Explanations (SHAP) were used to quantify the direction and magnitude of each predictor’s contribution to model predictions. Comparison with traditional statistical methods In addition to machine learning methods, weighted bivariate analyses with chi-square tests were performed in R to investigate the relationships between modern family planning use and explanatory variables. A multivariable logistic regression model was also fitted as a validation step to assess consistency between statistically significant predictors and those identified by machine learning models. Statistical Software and Implementation All machine learning studies were performed in Python (version 3.11), while R (version 4.4.3) was used for bivariate analysis, feature selection with the Boruta algorithm, geographic visualizations, and survey-weighted logistic regression. Using both software platforms enabled us to leverage Python’s strengths in machine learning and R’s capabilities in statistical analysis and feature selection, resulting in a robust, reproducible analytical framework. Results The socio-demographic and reproductive characteristics of married adolescent women aged 15–19 years, as well as their association with modern family planning (FP) use, are summarized in Table 1 . Overall, 72.66% (n = 2,342) of respondents reported utilizing a modern FP method, whereas 27.34% (n = 881) did not. The majority (57.1%) were between the ages of 18 and 19, while 75.7% had completed grades 6 through 10, and 93.8% were unemployed. Most participants had not been involved in FP programs during the previous three years (95.0%). Regarding marriage and having children, 85.3% were living with their husbands, 82.5% said they made FP decisions together, and 54.8% had given birth at some point. Media exposure was also different: almost half (49.4%) watched television daily, and 86.9% did not read newspapers or magazines. Most of them (77.1%) lived in rural areas, and the highest percentage of respondents (24.3%) was from the Dhaka division. As shown in Table 1 , chi-square tests revealed significant differences in modern FP use across several variables. FP use was higher among adolescents living with their husbands (77%), those sharing FP decisions with their partners (74.1%), and those who had ever given birth (80.7%) ( \(\:p\:\) < 0.001 for all). Exposure to mass media was also associated with FP use, women who watched television daily (74.5%) or read newspapers at least once a week (82.6%) showed higher use compared to those with no exposure ( \(\:p\:\) < 0.05). The regional difference was also significant (< 0.001), with the highest rates of modern FP usage in Mymensingh (80.1%) and Rajshahi (79%), followed by Rangpur (75.7%) and Barisal (73.9%), with the lowest in Chittagong (63.5%). Figure 1 shows a choropleth map of regional disparities in modern family planning (FP) use across Bangladesh, with darker shades indicating divisions with a higher prevalence of modern FP use. These spatial patterns highlight regional disparities in FP uptake, whereas no statistically significant differences were observed at the 5% level by age, education, employment, FP program participation, wealth, residence, or FP messages on television. Table 1 Background Characteristics of Married Adolescent Women and Bivariate Associations with Modern Family Planning Method Use, BAHWS 19–20. Characteristics \(\:\varvec{N}\) (%) Modern Family Planning Method Use Yes , \(\:\varvec{N}\) (%) No , \(\:\varvec{N}\) (%) \(\:\varvec{p}\) -value Age group 15–17 1382 (42.87) 1016 (73.5%) 365 (26.5%) 0.347 18–19 1841 (57.13) 1326 (72%) 516 (28%) Educational grades 0–5 601 (18.64) 456 (75.9%) 145 (24.1%) 0.066 6–10 2438 (75.65) 1761 (72.2%) 677 (27.8%) 11 and Higher 184 (5.70) 125 (67.9%) 59 (32.1%) Employment status No 3024 (93.82) 2204 (72.9%) 820 (27.1%) 0.294 Yes 199 (6.18) 138 (69.2%) 61 (30.8%) Involved in FP program in last 3 years No 3062 (95.01) 2219 (72.5%) 842 (27.5%) 0.392 Yes 161 (4.99) 122 (75.9%) 39 (24.1%) Currently residing with husband Living with her 2750 (85.34) 2117 (77%) 633 (23%) < 0.001 Staying elsewhere in Bangladesh 299 (9.27) 204 (68.1%) 95 (31.9%) Staying elsewhere outside Bangladesh 174 (5.39) 21 (12.2%) 153 (87.8%) Decision maker for FP use Both 2658 (82.48) 1971 (74.1%) 688 (25.9%) < 0.001 Mainly husband 275 (8.54) 166 (60.5%) 109 (39.5%) Mainly my decision 259 (8.02) 187 (72.2%) 72 (27.8%) Other 31 (0.96) 18 (58%) 13 (42%) Ever given birth No 1457 (45.22) 916 (62.9%) 541 (37.1%) < 0.001 Yes 1766 (54.78) 1426 (80.7%) 340 (19.3%) Wealth quantile Lowest 611 (18.95) 449 (73.6%) 161 (26.4%) 0.064 Second 711 (22.05) 541 (76.1%) 170 (23.9%) Middle 725 (22.50) 517 (71.3%) 208 (28.7%) Richer 686 (21.30) 497 (72.3%) 190 (27.7%) Highest 490 (15.20) 337 (68.8%) 153 (31.2%) Watch TV At least once a week 526 (16.31) 390 (74.1%) 136 (25.9%) 0.019 Every day 1591 (49.36) 1185 (74.5%) 406 (25.5%) Less than one a week 256 (7.95) 174 (67.9%) 82 (32.1%) Not at all 850 (26.38) 593 (69.7%) 257 (30.3%) Read Newspaper/Magazine At least once a week 128 (3.97) 101 (78.9%) 27 (21.1%) 0.008 Every day 20 (0.62) 16 (82.6%) 3 (17.4%) Less than one a week 276 (8.57) 179 (65%) 97 (35%) Not at all 2799 (86.85) 2045 (73.1%) 754 (26.9%) Heard about FP on TV (Last Few Months) No 2767 (85.86) 1999 (72.2%) 768 (27.8%) 0.197 Yes 456 (14.14) 343 (75.3%) 113 (24.7%) Read about FP in News (Last Few Months) No 3179 (98.64) 2303 (72.5%) 876 (27.5%) 0.040 Yes 44 (1.36) 38 (87.5%) 5 (12.5%) Residence Other urban 150 (4.67) 105 (69.6%) 46 (30.4%) 0.480 Rural 2484 (77.06) 1800 (72.5%) 684 (27.5%) Urban 589 (18.28) 437 (74.2%) 152 (25.8%) Administrative Division Barisal 214 (6.64) 158 (73.9%) 56 (26.1%) < 0.001 Chittagong 406 (12.58) 257 (63.5%) 148 (36.5%) Dhaka 783 (24.30) 548 (70%) 235 (30%) Khulna 438 (13.57) 303 (69.2%) 135 (30.8%) Mymensingh 304 (9.44) 244 (80.1%) 61 (19.9%) Rajshahi 569 (17.65) 450 (79%) 119 (21%) Rangpur 437 (13.57) 331 (75.7%) 106 (24.3%) Sylhet 72 (2.24) 51 (70.2%) 21 (29.8%) Total 3223 (100.0) 2342 (72.66) 881 (27.34) Feature Selection The Boruta algorithm was used to determine the most important features of modern contraceptive use among the original set of variables by iteratively comparing the feature importance scores with those of randomly permuted shadow features. This procedure approved features whose importance was greater than the maximum shadow value as significant and discarded those whose importance was less than the minimum or average shadow values. As shown in Fig. 2 , the Boruta analysis identified a set of key predictors that are significantly associated with modern FP use. The most influential features were currently residing with the husband, having ever given birth, age group, decision-making authority for FP use, and administrative division. Other variables, such as media exposure (TV and newspapers), wealth quantile, and participation in FP programs, also contributed moderately to model performance. Conversely, education, employment status, and place of residence had a small predictive power and were considered to be less important in further training of the model. Class Balancing Using SMOTE To balance the outcome variable, we used the Synthetic Minority Oversampling Technique (SMOTE) that creates synthetic observations in the minority class to balance the imbalance between the classes. Before balancing, the dataset had an unequal distribution, with 2,290 (72.34%) adolescents utilizing modern family planning (FP) methods and 876 (27.66%) not. After applying SMOTE, an equal number of observations were created for both categories, resulting in a symmetrical 50:50 distribution of modern FP users and non-users. Model Performance Comparison for predicting the Modern FP method use Several machine learning algorithms, such as Random Forest (RF), Decision Tree (DT), K-Nearest Neighbors (KNN), Logistic Regression (LR), Support Vector Machine (SVM), XGBoost, LightGBM, AdaBoost, Categorical Naïve Bayes, and Neural Networks, were used to develop a predictive model for modern FP method use among married adolescents in Bangladesh. Each model was trained using 70% of the data, with the remaining 30% for testing. Hyperparameter tuning was performed via GridSearchCV with 5-fold cross-validation to optimize model parameters which are provided in Supplementary Table 1. Model performance was evaluated using accuracy, precision, recall, F1-score, AUC (Area Under the Curve), Matthews Correlation Coefficient (MCC), and Cohen’s Kappa (Table 2 ). To address class imbalance, the evaluation included both imbalanced (original data) and balanced (SMOTE) scenarios, with XGBoost, Neural network, RF, and LR emerging as top performers. Models trained on the unbalanced data set achieved high overall accuracy but were evidently biased toward the majority group (FP users). The XGBoost and Random Forest models exhibited the highest accuracies (0.7811 each), accurately predicting approximately 78% of all cases. In the case of XGBoost, precision (0.7782), recall (0.9753), and F1-score (0.8656) were high and AUC of 0.7559 indicated good separation of classes, but the classification report indicated low recall of non-use (class 0: precision 0.81, recall 0.27, F1-score 0.41), reflecting poor performance of minority classes. MCC (0.3824) and Kappa (0.3129) of Cohen showed moderate reliability. RF also exhibited a similar pattern (accuracy = 0.7811; AUC = 0.7550), with very high recall among FP users (0.9884) but lower recognition among non-users. Table 2 Performance Metrics of Machine Learning Models for Modern FP Use Prediction with 95% Confidence Intervals, BAHWS 19–20. ML Models Peformance Unbalanced (Without SMOTE) Balanced (With SMOTE) Logistic Regression Accuracy 0.7716 ± 0.0274 0.7147 ± 0.0295 AUC 0.7537 ± 0.0334 0.7503 ± 0.0330 MCC 0.3519 ± 0.0610 0.3597 ± 0.0600 Cohen Kappa 0.2512 ± 0.0595 0.3522 ± 0.0600 Precision 0.7640 ± 0.0282 0.8455 ± 0.0282 Recall 0.9898 ± 0.0075 0.7409 ± 0.0345 F1-score 0.8624 ± 0.0182 0.7898 ± 0.0250 SVM Accuracy 0.7747 ± 0.0268 0.6611 ± 0.0305 AUC 0.6791 ± 0.0375 0.7285 ± 0.0343 MCC 0.3704 ± 0.0561 0.3271 ± 0.0585 Cohen Kappa 0.2558 ± 0.0584 0.3011 ± 0.0563 Precision 0.7642 ± 0.0281 0.8600 ± 0.0278 Recall 0.9956 ± 0.0050 0.6346 ± 0.0363 F1-score 0.8647 ± 0.0186 0.7303 ± 0.0280 KNN Accuracy 0.7611 ± 0.0268 0.7274 ± 0.0279 AUC 0.7242 ± 0.0354 0.7067 ± 0.0377 MCC 0.3030 ± 0.0647 0.2992 ± 0.0679 Cohen Kappa 0.2214 ± 0.0571 0.2985 ± 0.0682 Precision 0.7596 ± 0.0279 0.8006 ± 0.0295 Recall 0.9796 ± 0.0095 0.8297 ± 0.0277 F1-score 0.8557 ± 0.0185 0.8149 ± 0.0215 Neural Network Accuracy 0.7716 ± 0.0263 0.7358 ± 0.0279 AUC 0.7509 ± 0.0347 0.7453 ± 0.0353 MCC 0.3469 ± 0.0648 0.3708 ± 0.0605 Cohen Kappa 0.2670 ± 0.0609 0.3690 ± 0.0611 Precision 0.7683 ± 0.0270 0.8364 ± 0.0272 Recall 0.9796 ± 0.0108 0.7889 ± 0.0310 F1-score 0.8612 ± 0.0181 0.8120 ± 0.0219 Random Forest Accuracy 0.7811 ± 0.0268 0.7337 ± 0.0274 AUC 0.7550 ± 0.0350 0.7447 ± 0.0356 MCC 0.3878 ± 0.0608 0.3435 ± 0.0642 Cohen Kappa 0.2942 ± 0.0623 0.3434 ± 0.0643 Precision 0.7725 ± 0.0280 0.8210 ± 0.0266 Recall 0.9884 ± 0.0080 0.8079 ± 0.0286 F1-score 0.8672 ± 0.0184 0.8144 ± 0.0218 Decision Tree Accuracy 0.7737 ± 0.0269 0.7126 ± 0.0279 AUC 0.7283 ± 0.0338 0.6953 ± 0.0404 MCC 0.3568 ± 0.0601 0.2983 ± 0.0651 Cohen Kappa 0.2693 ± 0.0592 0.2979 ± 0.0653 Precision 0.7682 ± 0.0276 0.8099 ± 0.0283 Recall 0.9840 ± 0.0089 0.7875 ± 0.0296 F1-score 0.8628 ± 0.0183 0.7985 ± 0.0223 XGBoost Accuracy 0.7811 ± 0.0263 0.7600 ± 0.0263 AUC 0.7559 ± 0.0340 0.7285 ± 0.0379 MCC 0.3824 ± 0.0655 0.3720 ± 0.0663 Cohen Kappa 0.3129 ± 0.0639 0.3695 ± 0.0660 Precision 0.7782 ± 0.0279 0.8148 ± 0.0269 Recall 0.9753 ± 0.0115 0.8646 ± 0.0254 F1-score 0.8656 ± 0.0182 0.8390 ± 0.0201 LightGBM Accuracy 0.7642 ± 0.0268 0.7337 ± 0.0279 AUC 0.7120 ± 0.0384 0.6861 ± 0.0416 MCC 0.3316 ± 0.0722 0.3063 ± 0.0678 Cohen Kappa 0.3013 ± 0.0680 0.3047 ± 0.0680 Precision 0.7820 ± 0.0277 0.7997 ± 0.0285 Recall 0.9345 ± 0.0188 0.8428 ± 0.0259 F1-score 0.8515 ± 0.0192 0.8207 ± 0.0213 AdaBoost Accuracy 0.7726 ± 0.0258 0.7253 ± 0.0274 AUC 0.7581 ± 0.0340 0.7350 ± 0.0349 MCC 0.3500 ± 0.0653 0.3488 ± 0.0604 Cohen Kappa 0.2822 ± 0.0635 0.3468 ± 0.0604 Precision 0.7723 ± 0.0286 0.8307 ± 0.0280 Recall 0.9723 ± 0.0120 0.7787 ± 0.0318 F1-score 0.8608 ± 0.0180 0.8039 ± 0.0232 Categorical Naïve Bayes Accuracy 0.7716 ± 0.0268 0.6926 ± 0.0284 AUC 0.7377 ± 0.0343 0.7235 ± 0.0362 MCC 0.3460 ± 0.0654 0.2895 ± 0.0647 Cohen Kappa 0.2757 ± 0.0613 0.2860 ± 0.0638 Precision 0.7707 ± 0.0273 0.8170 ± 0.0296 Recall 0.9738 ± 0.0122 0.7409 ± 0.0317 F1-score 0.8605 ± 0.0180 0.7771 ± 0.0242 LightGBM showed slightly lower accuracy (0.7642) but maintained a reasonable AUC (0.7120) and F1-score (0.8515). The Neural Network (AUC = 0.7509) and Decision Tree (AUC = 0.7283) gave moderate results. Other less sophisticated models, such as LR (AUC = 0.7537), SVM (AUC = 0.6791), and KNN (AUC = 0.7242), did not perform well, with lower discriminative power and weaker agreement (e.g., MCC = 0.3519 for LR). Overall, models performed well on dominant patterns but struggled to capture minority (non-user) traits. Balancing the dataset with SMOTE increased sensitivity to the minority class, resulting in a more symmetric distribution of predictions. XGBoost performed best overall (accuracy = 0.7600; AUC = 0.7285), with high discrimination and stable agreement (MCC = 0.3720; κ = 0.3695). Its classification report demonstrated improved performance for non-users (precision 0.58, recall 0.49, F1 0.53) and consistent accuracy for users (precision 0.81, recall 0.86, F1 0.84), with an overall macro-F1 of 0.68 and weighted-F1 of 0.75. The Random Forest model also performed well (accuracy = 0.7337; AUC = 0.7447; MCC = 0.3435; k = 0.3434), with a high recall (0.8079) and F1-score (0.8144). The neural network achieved similar stability (accuracy = 0.7358; AUC = 0.7453; MCC = 0.3708), indicating reliable generalization under balanced conditions. AdaBoost and LightGBM demonstrated competitive performance (AUCs ≈ 0.69–0.74), while DT (AUC = 0.6953) and KNN (AUC = 0.7067) demonstrated moderate improvements. Simpler classifiers, LR (AUC = 0.7503) and SVM (AUC = 0.7285), had slightly higher Cohen’s Kappa values (0.3522 for LR), but their accuracies (0.7147 and 0.6611) were lower than ensemble counterparts. The Categorical Naïve Bayes model showed modest performance (AUC = 0.7235). Collectively, Fig. 3 shows that ensemble-based and deep learning models performed significantly better than standard classifiers. XGBoost received the highest overall composite score across all seven metrics, followed by Neural Network, Random Forest, and AdaBoost. These models consistently delivered greater recall, AUC, and MCC values, essential for imbalanced contexts, while retaining balanced accuracy. LightGBM and LR models provided adequate performance but slightly less stability. In contrast, DT, KNN, Naïve Bayes, and SVM were the weakest performers, characterized by lower Kappa (0.29–0.33) and MCC (< 0.31). Figure 4 shows the ROC curves for ten machine learning models evaluated on the SMOTE balanced test dataset. The ROC curve illustrates each model's ability to distinguish between users and non-users of modern family planning (FP) methods among married adolescent women. Among all models, Logistic Regression (AUC = 0.7503), Neural Network (AUC = 0.7453), and Random Forest (AUC = 0.7447) showed the most effective separation between the two groups, demonstrating a reasonable balance between sensitivity and specificity. AdaBoost (AUC = 0.7350), XGBoost and SVM (AUC = 0.7285) performed well, whereas Categorical Naïve Bayes provided moderate discrimination (AUC = 0.7235). KNN (AUC = 0.7067), Decision Tree (AUC = 0.6953), and LightGBM (AUC = 0.6861) performed less effectively, reflecting slightly weaker generalization to unseen data. Overall, ensemble and neural network models provided the most reliable classification results, while Logistic Regression, while being a simpler linear model, achieved one of the highest AUCs, demonstrating its strong interpretability and consistent predictive power. K-fold cross validation To ensure model stability and generalizability, a K-fold cross-validation process of 5, 10, 15, 20, 25, and 30 folds was used. Mean Accuracy (MAcc) and Precision–Recall AUC (PR-AUC) were calculated for each fold configuration across all models (Table 3 ). Across all folds, XGBoost, K-Nearest Neighbors (KNN), and Random Forest (RF) outperformed the other models. XGBoost maintained a high and consistent accuracy of 0.72–0.73, with a PR-AUC of 0.84–0.85, demonstrating strong discrimination between users and non-users of modern family planning methods. KNN showed high precision and stability across folds, with PR-AUC ranging from 0.82 to 0.83 and MAcc ranging from 0.71 to 0.72. With an MAcc of roughly 0.71 and a PR-AUC of 0.84–0.85, Random Forest also yielded consistent results, demonstrating robust, generalizable performance across fold sizes. Logistic Regression and Neural Network produced consistent results, with MAcc averaging 0.68 and PR-AUC ranging from 0.84 to 0.85 across all folds. AdaBoost demonstrated comparable stability, with MAcc around 0.70 and PR-AUC ranging from 0.84 to 0.85, indicating balanced and consistent predictive performance. LightGBM also performed well, with MAcc ranging between 0.70 and 0.71 and PR-AUC around 0.82, indicating efficient generalization to unseen data. Conversely, Decision Tree and Categorical Naïve Bayes had marginally inferior and less consistent performance, with Decision Tree achieving MAcc between 0.68 and 0.69 and PR-AUC between 0.80 and 0.82, whereas Naïve Bayes attained MAcc between 0.68 and 0.69 and PR-AUC between 0.83 and 0.84. The SVM showed the lowest accuracy, with a mean accuracy (MAcc) ranging from 0.63 to 0.64, while sustaining a PR-AUC of approximately 0.84, reflecting limited improvement across folds. Overall, the findings indicate that XGBoost, Random Forest and KNN are the most effective and stable models across all fold configurations. Table 3 K-fold cross-validation performance (Mean Accuracy and PR-AUC) of the selected machine learning models. Model 5 Folds 10 Folds 15 Folds 20 Folds 25 Folds 30 Folds MAcc PR-AUC MAcc PR-AUC MAcc PR-AUC MAcc PR-AUC MAcc PR-AUC MAcc PR-AUC Logistic Regression 0.68 0.84 0.68 0.84 0.68 0.84 0.68 0.84 0.68 0.84 0.68 0.85 SVM 0.63 0.84 0.64 0.84 0.64 0.84 0.64 0.84 0.64 0.84 0.64 0.84 KNN 0.72 0.82 0.72 0.82 0.72 0.82 0.72 0.82 0.71 0.83 0.72 0.83 Neural Network 0.68 0.84 0.68 0.84 0.68 0.84 0.68 0.83 0.68 0.84 0.68 0.84 Random Forest 0.71 0.84 0.71 0.84 0.71 0.84 0.71 0.84 0.71 0.84 0.71 0.85 Decision Tree 0.69 0.80 0.69 0.81 0.69 0.81 0.68 0.82 0.69 0.82 0.68 0.81 XGBoost 0.73 0.84 0.72 0.84 0.72 0.84 0.72 0.84 0.72 0.84 0.72 0.85 LightGBM 0.71 0.82 0.70 0.82 0.71 0.82 0.71 0.82 0.71 0.82 0.71 0.82 AdaBoost 0.70 0.84 0.70 0.84 0.70 0.85 0.70 0.84 0.70 0.84 0.70 0.85 Categorical Naïve Bayes 0.69 0.83 0.68 0.83 0.69 0.84 0.68 0.84 0.68 0.84 0.68 0.84 Feature Selection Results and Model Explainability Figure 5 displays the feature importance rankings obtained from the XGBoost model, highlighting the principal factors affecting modern family planning (FP) utilization among married adolescent women in Bangladesh. The variable “Currently residing with husband: Staying elsewhere outside Bangladesh” was the most significant predictor (importance = 0.218), indicating that women whose spouses reside overseas are considerably less likely to use modern family planning methods. The second most significant variable was “Ever given birth: No” (0.065), indicating that women who have not yet had childbirth often exhibit lower family planning utilization. Regional variation also played a substantial role, with Divisions such as Chittagong, Mymensingh, Rajshahi, and Dhaka exhibiting significant notable scores (ranging = 0.040–0.046), indicating persistent geographical disparities in FP access and uptake. Additionally, indicators of media exposure, particularly the frequency of reading newspapers or magazines weekly, were significant. The model’s predictions were also influenced by economic status variables, such as wealth quintiles (from lowest to wealthiest), and program-related factors (e.g., involvement in FP programs in the last 3 years and decision-making authority for FP use). These findings highlight the importance of both structural and relational factors in understanding variations in modern FP use among adolescent women. The integrated SHAP summary and feature importance plot (Fig. 6 ) shows the 20 most significant features identified by the XGBoost model for predicting modern family planning (FP) use. The features are ranked vertically by their mean absolute SHAP values, which indicate their average contribution to model predictions. Each dot represents an individual observation and is color-coded by feature value (red=high, blue = low), indicating how increases or decreases in that variable influence the probability of using modern FP. The strongest predictor was “Ever given birth: No” (mean |SHAP| = 0.449), with smaller SHAP values (blue) consistently reducing the predicted probability of modern FP use. This finding reflects that adolescent women who have not yet had a child are substantially less likely to adopt modern contraceptives. Conversely, women who have given birth (high feature values in red) were more likely to use FP methods. The second significant feature, "Currently residing with husband: Staying elsewhere outside Bangladesh" (mean |SHAP| = 0.294), had mostly negative SHAP values, indicating a lower likelihood of FP use when the husband resides abroad. The third related variable, "Currently residing with husband: Living with her," exhibited the opposite effect, contributing positively to the likelihood of FP use. Regional variables, including Chittagong, Rajshahi, Mymensingh, Khulna, and Dhaka divisions (0.034–0.105), exhibited mixed SHAP effects, reflecting both positive and negative influences depending on location. Notably, adolescents in Chittagong and Mymensingh divisions had higher SHAP variability, implying greater regional inequality in FP access and awareness. Socioeconomic status indicators, especially wealth quintiles (lowest to highest), exhibited a clear gradient. Wealth quintiles showed a clear directional gradient: higher wealth (red) was positively associated with modern FP use, while lower wealth (blue) was negatively associated, indicating a reduced probability of modern FP use. Media exposure variables, such as “Read any newspaper/magazine at least once a week,” had a positive impact on SHAP, whereas “Read not at all” had a negative impact. Behavioral and decision-related features, such as “Decision maker for FP use: Both” and “Involved in FP program in last 3 years: Yes,” exhibited positive SHAP values, suggesting increased probabilities of FP use. Conversely, their counterparts (“No” or “Mainly husband”) exhibited negative contributions. Overall, high SHAP values for childbirth history, shared FP decisions, and wealthier households indicate a greater likelihood of FP use, while negative SHAP values for spousal absence and poverty indicate reduced FP use. The fitted LR model in Supplementary Table 2 revealed that several explanatory variables remained statistically significant ( \(\:p\) < 0.05), aligning closely with the key predictors identified through SHAP analysis. Adolescents aged 18–19 years were significantly less likely to use modern FP methods compared to those aged 15–17 years (AOR = 0.71; 95% CI: 0.58–0.85; \(\:p\) = 0.0003), which aligns with the SHAP analysis results, also showing a positive value for the 15–17 age group. Participation in FP decision-making as “Both partners” (AOR = 1.61; 95% CI: 1.20–2.16; \(\:p\) = 0.0015) was positively associated with FP use, consistent with SHAP findings. Women who had ever given birth showed substantially higher odds of FP use (AOR = 3.20; 95% CI: 2.62–3.89; \(\:p\) < 0.0001), reaffirming its dominant importance from the SHAP model. Those whose husbands were staying elsewhere in Bangladesh (AOR = 0.74, \(\:p\) = 0.044) or outside the country (AOR = 0.03, \(\:p\) < 0.001) had lower odds, matching SHAP’s strong negative effects for spousal absence. Reading newspapers at least once a week (AOR = 1.72, \(\:p\) = 0.028) and joint decision-making with the husband (AOR = 1.61, \(\:\:p\) = 0.002) were associated with higher FP use, consistent with positive SHAP contributions. Similarly, wealthier households (AOR range = 1.38–1.65, \(\:p\) < 0.05) and residence in Mymensingh ( \(\:p\) = 0.027) and Rajshahi ( \(\:p\) = 0.023) divisions showed positive effects, paralleling SHAP findings of regional and socioeconomic influence. Discussion The modern family planning (FP) methods are necessary to empower married adolescent women so that they can have control over their reproductive health, especially in Bangladesh, where early marriage is still a common practice and causes high levels of unwanted pregnancies and maternal deaths [ 11 , 12 ]. This study used a comprehensive machine learning (ML) framework to predict modern family planning (FP) use among married adolescent women aged 15–19 years in Bangladesh using nationally representative BAHWS 2019-20 data. The prevalence of modern FP use was 72.66%, with significant geographical differences: Mymensingh and Rajshahi had the highest rates, while Chittagong had the lowest. Ensemble methods were more effective in prediction across models, particularly when SMOTE was used to mitigate class imbalance. XGBoost turned out to be the best-performing model, with improved accuracy, AUC, precision, and recall compared to other models. The most important determinants were spousal co-residency (husband living abroad and husband living with her), parity (ever having a baby), administrative division, FP decision-making dynamics, age group, media exposure, and wealth quantile, with strong consistency between SHAP analysis and survey-weighted logistic regression. The prevalence of modern FP use in our adolescent sample exceeds the national adolescent contraceptive prevalence rate from the BDHS summary findings [ 8 ], possibly due to our focus on married current users and survey-specific sampling. However, it remains lower than among older women, highlighting disparities driven by adolescents’ limited autonomy and relational barriers [ 13 , 37 ]. Regional clustering, visualized in Fig. 1 , aligns with BDHS analyses showing southeastern divisions like Chittagong facing service inequities and conservative norms [ 38 ]. The finding is higher than in high-fertility Sub-Saharan African countries[ 24 ] but lower than in urban Kenya [ 39 ], reflecting Bangladesh’s progress in FP programs amid persistent adolescent gaps. Ten ML algorithms were trained on balanced and unbalanced datasets, with performance evaluated using accuracy, AUC, precision, recall, F1-score, Matthews Correlation Coefficient (MCC), and Cohen’s Kappa. In the unbalanced dataset, several models achieved relatively high overall accuracy but showed poor sensitivity to the minority class (non-users), indicating majority-class bias. This is a common issue in health prediction problems where the outcome distribution is asymmetric, and it highlights why relying only on accuracy can be misleading [ 40 ]. After applying SMOTE, performance became more balanced, and ensemble models remained superior. XGBoost achieved the best overall composite performance across the evaluated metrics and also showed strong stability in k-fold cross-validation, a finding which aligns with a similar study on modern FP use among reproductive-age women in Ethiopia, where the XGBoost model outperformed other classifiers [ 41 ]. Our findings are consistent with evidence from other ML studies, in which XGBoost and Random Forest often outperform conventional classifiers in predicting FP-related outcomes such as modern FP non-use, optimal ANC utilization, or fertility preferences [ 3 , 42 , 43 ]. SMOTE addressed class imbalance by improving recall among the minority class (non-users), which is consistent with improvements in Ethiopian fertility models [ 29 ]. This robustness, with ensemble methods outperforming simpler classifiers such as Decision Tree (DT) or K-Nearest Neighbors (KNN), demonstrates machine learning's shift from linear assumptions in chi-square tests to capturing complex interactions [ 44 ]. Unlike logistic regression, which may fail due to multicollinearity and other statistical violations, ensemble methods effectively capture the complex, nonlinear interactions found in public health data, making them more reliable for prediction in this study [ 45 ]. Spousal co-residency emerged as the most influential determinant across Boruta selection, XGBoost importance, and SHAP analyses. Adolescents whose husbands lived outside Bangladesh had extremely low FP use, which is consistent with studies in Bangladesh [ 46 ] and India [ 47 ] that found reduced contraceptive uptake among women experiencing spousal separation due to migration. In contrast, research conducted in East Africa that employs machine learning methods identifies marital status (married vs. unmarried) as a significant predictor, rather than spousal co-residency [ 48 ]. This discrepancy is likely due to contextual differences in Bangladesh, where nearly all adolescents in the sample were married; spousal co-residency serves as a more informative marker of FP use than marital status alone. Adolescents residing with their husbands were significantly more likely to use modern family planning, a finding that was consistently prioritized across Boruta selection, XGBoost importance, and SHAP analyses. This is consistent with patterns in South Asia, where co-residency with husbands and, in some cases, with other family members such as mothers-in-law, is associated with higher use of modern contraceptive methods [ 49 ]. Adolescents who had ever given birth were substantially more likely to use modern FP, consistent with previous Bangladesh studies showing that contraceptive adoption often follows the first birth rather than preceding it [ 50 , 51 ]. Joint decision-making for FP use showed a positive association with FP uptake, aligning with prior findings from Bangladesh and South Asia that emphasize the role of couple communication and shared authority [ 52 – 54 ]. This result is consistent with an ML-based FP study in East Africa that highlights women’s autonomy and decision-making power as key predictors, even though it modeled the non-use of family planning, unlike this study’s focus on FP uptake [ 3 ]. Socioeconomic status also played an important role in modern FP use. Adolescents from wealthier households were more likely to use modern FP methods, while those in the lowest wealth quintile had lower uptake. This finding is consistent with previous studies in Bangladesh and Nigeria that demonstrate that economic disadvantage limits access to contraceptive services and information, thus reducing utilization [ 55 , 56 ]. Although the bivariate association between wealth and FP use was marginal in this study, the SHAP results suggest that wealth interacts with other factors, such as geographic location and media exposure, in complex ways that may not be fully captured by traditional statistical tests. Exposure to media, particularly reading newspapers or magazines, was positively associated with the use of modern FP. Adolescents with no exposure to newspapers or magazines were less likely to use FP, whereas those with regular exposure had higher uptake. This aligns with prior evidence in Ethiopia, indicating that access to information through TV, radio, and magazines increases awareness of modern contraceptive options and improves informed decision-making [ 57 ]. Age differences were also observed in modern FP use among married adolescents. Although the chi-square test indicated no significant bivariate association between age group and modern FP use, Boruta feature selection and SHAP analysis identified age group as an important predictor, with the 15–17-year age group contributing positively to the predicted FP probability compared to 18–19-year-olds. Adolescents aged 15–17 years were more likely to use modern FP methods than those aged 18–19 years, after adjustment for other factors. This aligns with a systematic review in Bangladesh that found younger women utilize contraceptives more frequently than older women [ 58 ]. It also corresponds with DHS evidence from 2017–18, which indicates that younger contraceptive users (ages 15–24) are more likely to obtain methods from the private sector than older users [ 59 ]. Geographic variation in FP use remained pronounced in this study. Adolescents residing in Mymensingh and Rajshahi divisions had higher modern FP use than those in Chittagong, which consistently showed lower uptake. These findings are consistent with prior Bangladesh research showing persistent geographic inequalities in FP uptake across divisions, reflecting differences in service delivery, program intensity, infrastructure, and sociocultural norms [ 60 , 61 ]. Participation in FP programs over the past three years had a modest effect on FP use. While program involvement contributed positively in the SHAP analyses, it did not emerge as a strong independent predictor in conventional regression models. The integration of SHAP and Boruta allowed this study to move beyond prediction to interpretation. Similar to a recent ML study on modern FP outcome [ 3 ] SHAP revealed directional effects and interaction patterns not easily captured by traditional regression. The close alignment between SHAP-derived important predictors and survey-weighted logistic regression results supports the validity of the ML framework. Strengths, limitations, and future directions This study’s strengths include the use of nationally representative adolescent data, applying major data preprocessing, systematic comparison of multiple ML models, explicit handling of class imbalance, and integration of SHAP analysis with regression validation. Limitations include the cross-sectional design, reliance on self-reported FP use, exclusion of unmarried adolescents, limited data on proximal determinants, and lack of external validation. Future research should incorporate longitudinal data, include proximal factors like autonomy and service quality, and apply deep learning. Conclusion This study applied and compared multiple machine learning classification models to predict modern family planning (FP) use and identify its key determinants among married adolescent women aged 15–19 years in Bangladesh using nationally representative BAHWS 2019–20 data. We used various machine learning classifiers, such as Logistic Regression, SVM, KNN, Decision Tree, Random Forest, Neural Network, and XGBoost. Ensemble-based approaches outperformed conventional classifiers, with XGBoost demonstrating the strongest and most stable predictive performance, achieving an accuracy of 76%, precision of 81.5%, recall of 86.5%, and an AUC-ROC of 72.9% after addressing class imbalance with SMOTE. The findings indicate that modern FP use among married adolescents is primarily influenced by spousal co-residency, childbirth experience, decision-making authority for family planning, administrative division, age group, wealth status, and media exposure. Adolescents whose husbands were residing outside Bangladesh and those who had not yet given birth were substantially less likely to use modern FP methods. This study also revealed that adolescent girls from the poorest households and those with no exposure to newspapers or magazines were less likely to use modern FP methods. In contrast, adolescents who made joint family planning decisions with their husbands, belonged to wealthier households, and were exposed to newspapers or magazines were more likely to use modern family planning. In contrast, adolescents who engaged in joint decision-making with their husbands, were in the younger age group (15–17 years), and belonged to wealthier households, were significantly more likely to use modern FP. Geographic disparities were also evident, with higher FP use observed in Mymensingh and Rajshahi and lower use in Chittagong, Khulna, and Dhaka. The consistency among machine learning, SHAP analysis, and multivariable logistic regression strengthens the validity of these findings and demonstrates that ensemble machine learning models, particularly XGBoost, are effective at identifying specific risk factors among adolescents using modern family planning and provide valuable evidence for targeted interventions. Policymakers and program planners should consider these findings when designing adolescent-focused family planning strategies, with particular attention to adolescents affected by spousal separation, limited decision-making power, and regional and socioeconomic inequalities. Strengthening targeted approaches is essential to expanding equitable access to modern family planning services and achieving Sustainable Development Goal (SDG) target 3.7 in Bangladesh. Declarations Ethics approval and consent to participate This study used secondary, publicly available data. Ethical approval for the original survey was obtained from the relevant authorities. Therefore, additional ethical approval and informed consent were not required for this study. Consent for publication Not applicable. Availability of data and materials The data utilized in the current research can be accessed without registration at the UNC Dataverse (https://dataverse.unc.edu/). Competing interests The authors have declared that no competing interests exist. Funding This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Authors’ contributions Conceptualization: Shawkatul Islam, Muhammad Khairul Alam Data curation: Shawkatul Islam Formal analysis: Shawkatul Islam Methodology: Shawkatul Islam, Md Eyah Eya, Muhammad Khairul Alam Software: Md. Rayhan Kabir, Md Eyah Eya Supervision: Muhammad Khairul Alam Validation: Muhammad Khairul Alam Visualization: Shawkatul Islam Writing – original draft: Shawkatul Islam, Md Eyah Eya Writing – review & editing: Muhammad Khairul Alam All authors approved the final version of the manuscript. Acknowledgments The authors are grateful to the National Institute of Population Research and Training (NIPORT), the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b), and Data for Impact (D4I) for their assistance in carrying out the Bangladesh Adolescent Health and Wellbeing Survey (BAHWS) 2019-20. We also like to thank the Bangladeshi government, the United States Agency for International Development (USAID)/Bangladesh, and the UK Foreign, Commonwealth, and Development Office (FCDO) for their assistance with the survey and for making the dataset publicly available. Authors’ information Not applicable. References Festin MPR. Overview of modern contraception. Best Pract Res Clin Obstet Gynaecol. 2020;66: 4–14. doi:10.1016/j.bpobgyn.2020.03.004 Tsui AO, McDonald-Mosley R, Burke AE. Family planning and the burden of unintended pregnancies. Epidemiol Rev. 2010;32: 152–174. doi:10.1093/EPIREV/MXQ012 Yehuala TZ. Exploring machine learning algorithms to predict not using modern family planning methods among reproductive age women in East Africa. BMC Health Services Research 2024 24:1. 2024;24: 1595-. doi:10.1186/S12913-024-11932-X JR B, BJ S, AO L. Improving Birth Outcomes: Meeting the Challenge in the Developing World. Improving Birth Outcomes. 2003 [cited 31 Jan 2026]. doi:10.17226/10841 Family planning/contraception methods. [cited 31 Jan 2026]. Available: https://www.who.int/news-room/fact-sheets/detail/family-planning-contraception Jimmy E, OsonwaKalu O, Nelson O, Dominic O. Prevalence of Contraceptive use among women of reproductive age in Calabar Metropolis, Southern Nigeria. 2013. World Family Planning 2022 Meeting the changing needs for family planning: Contraceptive use by age and method. NIPORT. National Institute of Population Research and Training (NIPORT), Mitra and Associates, & ICF International. (2023). Bangladesh Demographic and Health Survey 2022–23: Key Indicators Report. Dhaka, Bangladesh, and Rockville, Maryland, USA: NIPORT, Mitra and Associates, and ICF International. 2023. Islam AZ, Rahman M, Mostofa MG. Association between contraceptive use and socio-demographic factors of young fecund women in Bangladesh. Sex Reprod Healthc. 2017;13: 1–7. doi:10.1016/J.SRHC.2017.05.001 Population Reference Bureau. World population data sheet. Population Reference Bureau Washington. 2015. . 2015. Kamal N. Contraceptive use among married adolescent girls in Bangladesh. J Biosoc Sci. 2013;45: 71–86. Shahabuddin ASM, Nöstlinger C, Delvaux T, Sarker M, Bardají A, Brouwere V De, et al. What Influences Adolescent Girls’ Decision-Making Regarding Contraceptive Methods Use and Childbearing? A Qualitative Exploratory Study in Rangpur District, Bangladesh. PLoS One. 2016;11: e0157664. doi:10.1371/journal.pone.0157664 Kamrul Islam M, Rabiul Haque M, Hema PS. Regional variations of contraceptive use in Bangladesh: A disaggregate analysis by place of residence. PLoS One. 2020;15. doi:10.1371/journal.pone.0230143 Islam AZ, others. Association between contraceptive use and socio-demographic factors of young fecund women in Bangladesh. Sex Reprod Healthc. 2017;13: 88–94. SDG Target 3.7 Sexual and reproductive health. [cited 31 Jan 2026]. Available: https://www.who.int/data/gho/data/themes/topics/sdg-target-3_7-sexual-and-reproductive-health Haq I, Sakib S, Talukder A. Sociodemographic Factors on Contraceptive Use among Ever-Married Women of Reproductive Age: Evidence from Three Demographic and Health Surveys in Bangladesh. Medical Sciences. 2017;5: 31. doi:10.3390/medsci5040031 Hossain M, Khan M, Ababneh F, Shaw J. Identifying factors influencing contraceptive use in Bangladesh: Evidence from BDHS 2014 data. BMC Public Health. 2018;18. doi:10.1186/s12889-018-5098-1 Kundu S, Kundu S, Rahman MA, Kabir H, Al Banna MH, Basu S, et al. Prevalence and determinants of contraceptive method use among Bangladeshi women of reproductive age: a multilevel multinomial analysis. BMC Public Health. 2022;22. doi:10.1186/s12889-022-14857-4 Hossain MI, others. Performance evaluation of machine learning algorithm for classification of unintended pregnancy among married women in Bangladesh. J Healthc Eng. 2022;2022: 1460908. Sidey-Gibbons JAM, Sidey-Gibbons CJ. Machine learning in medicine: a practical introduction. BMC Med Res Methodol. 2019;19: 64. doi:10.1186/s12874-019-0681-4 Kebede SD, Sebastian Y, Yeneneh A, Chanie AF, Melaku MS, Walle AD. Prediction of contraceptive discontinuation among reproductive-age women in Ethiopia using Ethiopian Demographic and Health Survey 2016 Dataset: A Machine Learning Approach. BMC Med Inform Decis Mak. 2023;23. doi:10.1186/s12911-023-02102-w Walle AD, Kebede SD, Adem JB, Mamo DN. Machine-learning algorithm to predict home delivery after antenatal care visit among reproductive age women in East Africa. Front Glob Womens Health. 2025;6. doi:10.3389/fgwh.2025.1461475 Yehuala TZ. Exploring machine learning algorithms to predict not using modern family planning methods among reproductive age women in East Africa. BMC Health Serv Res. 2024;24: 1595. Melaku MS, Yohannes L, Sharew B, Derseh MH, Taye EA. Application of machine learning algorithms to model predictors of informed contraceptive choice among reproductive age women in six high fertility rate sub Sahara Africa countries. BMC Public Health. 2025;25. doi:10.1186/s12889-025-23242-w Hossain MI, Habib MJ, Saleheen AAS, Kamruzzaman M, Rahman A, Roy S, et al. Performance Evaluation of Machine Learning Algorithm for Classification of Unintended Pregnancy among Married Women in Bangladesh. J Healthc Eng. 2022;2022. doi:10.1155/2022/1460908 Chen T, Guestrin C. XGBoost: A scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2016. pp. 785–794. doi:10.1145/2939672.2939785 Breiman L. Random Forests. 2001. Gupta S, Saluja K, Goyal A, Vajpayee A, Tiwari V. Comparing the performance of machine learning algorithms using estimated accuracy. Measurement: Sensors. 2022;24. doi:10.1016/j.measen.2022.100432 Sani J, Halane S, Ahmed AM, Ahmed MM. Application of machine learning algorithms and SHAP explanations to predict fertility preference among reproductive women in Somalia. Sci Rep. 2025;15. doi:10.1038/s41598-025-04704-y (NIPORT) NI of PR and T, International Centre for Diarrhoeal Disease Research b) B (icddr, Impact D for. Bangladesh Adolescent Health and Wellbeing Survey 2019-20. NIPORT, icddr, b, and Data for Impact Dhaka, Bangladesh, and Chapel Hill, NC …; 2021. World Health Organization. Reproductive Health and Research, World Health Organization. Medical eligibility criteria for contraceptive use, 5th ed. 2015: Geneva. 2015; 276. Kamrul Islam M, Rabiul Haque M, Hema PS. Regional variations of contraceptive use in Bangladesh: A disaggregate analysis by place of residence. PLoS One. 2020;15: e0230143. doi:10.1371/JOURNAL.PONE.0230143 AZ I, MN M, ML K, MM R, MR I, MG M, et al. Prevalence and Determinants of Contraceptive use among Employed and Unemployed Women in Bangladesh. Int J MCH AIDS. 2016;5. doi:10.21106/IJMA.83 Rana MS, Khanam SJ, Alam MB, Hassen MT, Kabir MI, Khan MN. Exploration of modern contraceptive methods using patterns among later reproductive-aged women in Bangladesh. PLoS One. 2024;19: e0291100. doi:10.1371/JOURNAL.PONE.0291100 Yang S, Berdine G. The receiver operating characteristic (ROC) curve. 2017;5: 34–36. doi:10.12746/SWRCCC.V5I19.391 Kursa MB, Rudnicki WR. Feature Selection with the Boruta Package. J Stat Softw. 2010;36: 1–13. doi:10.18637/JSS.V036.I11 Mahmud M, Islam MM. Adolescent contraceptive use and its determinants in Bangladesh: Evidence from Bangladesh Fertility Survey 1989. Contraception. 1995;52: 181–186. doi:10.1016/0010-7824(95)00149-5 Huda F, Chowdhuri S, Sarker B, Islam N, Ahmed A. Prevalence of unintended pregnancy and needs for family planning among married adolescent girls living in urban slums of Dhaka, Bangladesh. 2014. doi:10.31899/rh4.1050 Feeser K, Chakraborty NM, Calhoun L, Speizer IS. Measures of family planning service quality associated with contraceptive discontinuation: an analysis of Measurement, Learning & Evaluation (MLE) project data from urban Kenya. Gates Open Res. 2020;3: 1453. doi:10.12688/gatesopenres.12974.2 Van Den Goorbergh R, Van Smeden M, Timmerman D, Ben Van Calster. The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression. Journal of the American Medical Informatics Association. 2022;29: 1525–1534. doi:10.1093/JAMIA/OCAC093 Adem JB, Kebede SD, Walle AD, Mamo DN. Predicting determinants of modern contraceptive use among reproductive-age women in Ethiopia using machine learning algorithm: Evidence from the Performance Monitoring and Accountability (PMA) Survey 2019 dataset. F1000Research 2025 14:99. 2025;14: 99. doi:10.12688/f1000research.156316.1 Sani J, Oluwagbemiga A, Ahmed MM. Machine learning-based prediction of optimal antenatal care utilization among reproductive women in Nigeria. Machine Learning with Applications. 2025;21: 100698. doi:10.1016/J.MLWA.2025.100698 Tadese ZB, Nimani TD, Mare KU, Gubena F, Wali IG, Sani J. Exploring machine learning algorithms for predicting fertility preferences among reproductive age women in Nigeria. Front Digit Health. 2025;6: 1495382. doi:10.3389/FDGTH.2024.1495382 Lundberg S, Lee S-I. A Unified Approach to Interpreting Model Predictions. 2017. Available: http://arxiv.org/abs/1705.07874 Rois R, Ray M, Rahman A, Roy SK. Prevalence and predicting factors of perceived stress among Bangladeshi university students using machine learning algorithms. Journal of Health, Population and Nutrition 2021 40:1. 2021;40: 50-. doi:10.1186/S41043-021-00276-5 Khan R, MacQuarrie KLD, Sultana M, Nahar Q. Intermittent Needs for Family Planning among Women with an Internal Migrant Husband in Bangladesh: A Qualitative Study. Sex Reprod Health Matters. 2022;29: 2097044. doi:10.1080/26410397.2022.2097044 Samanta R, Munda J. Husband’s migration status and contraceptive behaviors of women: evidence from Middle-Ganga Plain of India. BMC Womens Health. 2023;23: 180. doi:10.1186/S12905-023-02325-Z Tesfa GA, Demeke AD, Seboka BT, Tebeje TM, Kasaye MD, Gebremeskele BT, et al. Employing machine learning models to predict pregnancy termination among adolescent and young women aged 15–24 years in East Africa. Scientific Reports 2024 14:1. 2024;14: 30047-. doi:10.1038/s41598-024-81197-1 Pradhan MR, Mondal S. Examining the influence of Mother-in-law on family planning use in South Asia: insights from Bangladesh, India, Nepal, and Pakistan. BMC Women’s Health 2023 23:1. 2023;23: 418-. doi:10.1186/s12905-023-02587-7 Khan MN, Khan MMA, Billah MA, Khanam SJ, Haider MM, Sarker BK, et al. Effects of maternal healthcare service utilization on modern postpartum family planning access in Bangladesh: insights from a National representative survey. PLoS One. 2025;20: e0318363. doi:10.1371/JOURNAL.PONE.0318363 Islam MM, Islam MK, Hasan MS, Hossain MB. Adolescent motherhood in Bangladesh: Trends and determinants. PLoS One. 2017;12: e0188294. doi:10.1371/JOURNAL.PONE.0188294 Rahman MM, Mostofa MG, Hoque MA. Women’s household decision-making autonomy and contraceptive behavior among Bangladeshi women. Sexual & Reproductive Healthcare. 2014;5: 9–15. doi:10.1016/J.SRHC.2013.12.003 Hameed W, Azmat SK, Ali M, Sheikh MI, Abbas G, Temmerman M, et al. Women’s Empowerment and Contraceptive Use: The Role of Independent versus Couples’ Decision-Making, from a Lower Middle Income Country Perspective. PLoS One. 2014;9: e104633. doi:10.1371/JOURNAL.PONE.0104633 Islam AZ. Factors affecting modern contraceptive use among fecund young women in Bangladesh: does couples’ joint participation in household decision making matter? Reprod Health. 2018;15: 112. doi:10.1186/S12978-018-0558-8 Alam N, Mollah MMH, Naomi SS. Prevalence and determinants of adolescent childbearing: comparative analysis of 2017–18 and 2014 Bangladesh Demographic Health Survey. Front Public Health. 2023;11: 1088465. doi:10.3389/FPUBH.2023.1088465/FULL Akinyemi AI, Ikuteyijo OO, Mobolaji JW, Erinfolami T, Adebayo SO. Socioeconomic inequalities and family planning utilization among female adolescents in urban slums in Nigeria. Front Glob Womens Health. 2022;3: 838977. doi:10.3389/FGWH.2022.838977 Yesuf KA, Liyew AD, Bezabih AK. Impact of exposure to mass media on utilization modern contraceptive among adolescent married women in Ethiopia: evidence from Ethiopia demographic health survey 2016. International Journal of Scientific Reports. 2021;7: 434. doi:10.18203/issn.2454-2156.IntJSciRep20213257 Moon MP. Contraceptive behaviors and media influence among women in Bangladesh: exploring the effects of age and education. Front Glob Womens Health. 2025;6: 1492105. doi:10.3389/fgwh.2025.1492105 Plus S. Sources of Family Planning Bangladesh. 2016. Khan MHR, Siddik AB, Islam T. Disparities in contraceptive preferences among Bangladeshi women: a multilevel logistic regression study. BMC Public Health. 2025;25: 3444. doi:10.1186/s12889-025-24424-2 Kamrul Islam M, Rabiul Haque M, Hema PS. Regional variations of contraceptive use in Bangladesh: A disaggregate analysis by place of residence. PLoS One. 2020;15: e0230143. doi:10.1371/journal.pone.0230143 Additional Declarations No competing interests reported. Supplementary Files SupplementaryTables.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 30 Mar, 2026 Reviewers agreed at journal 15 Mar, 2026 Reviews received at journal 14 Mar, 2026 Reviews received at journal 13 Mar, 2026 Reviewers agreed at journal 13 Mar, 2026 Reviewers agreed at journal 11 Mar, 2026 Reviewers agreed at journal 08 Mar, 2026 Reviewers agreed at journal 06 Mar, 2026 Reviewers invited by journal 06 Mar, 2026 Editor assigned by journal 04 Mar, 2026 Editor invited by journal 11 Feb, 2026 Submission checks completed at journal 11 Feb, 2026 First submitted to journal 11 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8806294","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":604571721,"identity":"906a9691-a0c2-41e1-8873-bada901ffaa1","order_by":0,"name":"Shawkatul Islam","email":"","orcid":"","institution":"Jahangirnagar University","correspondingAuthor":false,"prefix":"","firstName":"Shawkatul","middleName":"","lastName":"Islam","suffix":""},{"id":604571722,"identity":"7f5c4197-60aa-4bbe-8dd0-d5691f668d39","order_by":1,"name":"Md Eyah Eya","email":"","orcid":"","institution":"Jahangirnagar University","correspondingAuthor":false,"prefix":"","firstName":"Md","middleName":"Eyah","lastName":"Eya","suffix":""},{"id":604571724,"identity":"50996e9f-4b6b-4104-913c-9002d5cd9ef3","order_by":2,"name":"Muhammad Khairul Alam","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1ElEQVRIie3NMQrCMBSA4QcFu4RmTZf2CpFCXIpneUXwCEUQbEHQK+gtdOkcCdilB6h00cErSN00KroZ3QTzDy9veB8BsNl+MffxUpAAqDdpIo4eCMzPvyb8eWkidOoc1+e2P4k2ldgfZhB4Nb4nTHVEQ3DAhKx6PJlB5JsIKBANoKOJYFhBsjKRULmnXYsZi/I7yYyEKyJqgopx0GQEyE2kq0jakGHpL+Q2vRLWXVb79yQo58WujceULlThtzwOvdLwyysmb/PTcx3Nv7m22Wy2f+oCP6VG89ezZPIAAAAASUVORK5CYII=","orcid":"","institution":"Jahangirnagar University","correspondingAuthor":true,"prefix":"","firstName":"Muhammad","middleName":"Khairul","lastName":"Alam","suffix":""}],"badges":[],"createdAt":"2026-02-06 11:23:30","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8806294/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8806294/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104514681,"identity":"a271a308-1dee-4693-b58d-4f2fa107a24e","added_by":"auto","created_at":"2026-03-12 17:06:43","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":153349,"visible":true,"origin":"","legend":"\u003cp\u003eGeographical distribution of modern family planning method use among married adolescent women in Bangladesh.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8806294/v1/a1620e54842ae87ead674047.png"},{"id":104514679,"identity":"97510814-e499-4437-a5de-102ad4b09d25","added_by":"auto","created_at":"2026-03-12 17:06:43","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":65353,"visible":true,"origin":"","legend":"\u003cp\u003eFeature selection using the Boruta algorithm.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8806294/v1/3855762d1a15b3b76c2a72a3.png"},{"id":104514684,"identity":"fb6122f9-1bb3-40b6-a5ea-dd83bf1f0027","added_by":"auto","created_at":"2026-03-12 17:06:43","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":95982,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of model performance metrics (accuracy, precision, recall, F1-score, AUC, MCC, and Kappa) for SMOTE-balanced data.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8806294/v1/72814cc4e16b5de0ee36e97a.png"},{"id":104514685,"identity":"04dc7a21-9233-4249-a839-dd865e08210d","added_by":"auto","created_at":"2026-03-12 17:06:43","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":165011,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves of selected machine learning models for predicting modern family planning use.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8806294/v1/57787f94cda67f83a451c9ae.png"},{"id":104514680,"identity":"5a9e764d-d40f-4466-8b7f-1620dfd75b96","added_by":"auto","created_at":"2026-03-12 17:06:43","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":114041,"visible":true,"origin":"","legend":"\u003cp\u003eImportant determinants selected by XGBoost model.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8806294/v1/842009f6e54ab41a498edb96.png"},{"id":104780806,"identity":"ad3a1786-b7dd-4df1-abf7-052e59b69163","added_by":"auto","created_at":"2026-03-17 07:54:00","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":141383,"visible":true,"origin":"","legend":"\u003cp\u003eCombined feature importance (left) and SHAP summary (right) plots from the XGBoost model showing the top 20 predictors.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-8806294/v1/7186c985aac3f2663787d518.png"},{"id":104835117,"identity":"5d4df867-92ed-4b05-a464-89030d58a109","added_by":"auto","created_at":"2026-03-17 17:40:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2656665,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8806294/v1/a4c7e5ef-65b3-4808-9841-a439eb5fc5eb.pdf"},{"id":104514682,"identity":"81bddc97-56fa-40f7-ac60-f9f5295f9383","added_by":"auto","created_at":"2026-03-12 17:06:43","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":28651,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-8806294/v1/6749f366a18e5d962618d171.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Determinants of modern family planning use among married adolescents in Bangladesh: Evidence from a machine learning approach","fulltext":[{"header":"Introduction","content":"\u003cp\u003eModern family planning methods encompass a variety of contraceptive methods, including the use of oral pills, injectables, intrauterine devices (IUDs), implants, condoms, emergency contraception, and permanent forms of contraception, such as female and male sterilization [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The widespread use and accessibility of these methods are crucial for preventing unwanted pregnancies, reducing maternal, perinatal, and neonatal mortality, and facilitating effective birth spacing [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Beyond their direct health benefits, modern contraceptive methods contribute to broader social gains by giving women enough time to recover between pregnancies, improving child health and educational outcomes, and contributing to overall socioeconomic growth [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAmong the 1.9\u0026nbsp;billion women of reproductive age (15\u0026ndash;49 years) worldwide in 2021, an estimated 1.1\u0026nbsp;billion had a need for family planning; of these, 874\u0026nbsp;million were using modern contraceptive methods, while 164\u0026nbsp;million had an unmet need for contraception [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. Projections indicate that by 2030, approximately 1.19\u0026nbsp;billion women will require modern family planning methods. However, the use of modern contraception varies across countries and regions [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. The use of modern contraception among married women of reproductive age is considerably lower in developing nations (45%) than in developed countries (70%), with notable regional variation, including 56% in Latin America, 74% in East Asia, 34% in South Asia, and 14% in Africa [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. The particularly low prevalence of modern contraceptive use in South Asia is concerning, given that Bangladesh is part of this region and continues to experience high levels of early marriage. On a global and regional scale, women under the age of 25 consistently demonstrate the lowest levels of contraceptive use and pregnancy avoidance in comparison to women aged 25\u0026ndash;44 years. This highlights a huge unmet need among adolescents and young women [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eBangladesh has a large and young population, with nearly half of its 160.4\u0026nbsp;million people below the age of 25, and it continues to report one of the highest rates of child marriage globally [\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. As an increasing number of adolescents enter marriage each year, early marriage exposes young women to frequent and often unprotected sexual intercourse, thereby increasing the risk of early, rapid, and unintended pregnancies with adverse maternal and neonatal health consequences [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. While Bangladesh has achieved significant strides in increasing the use of contraceptives on a national level, these improvements conceal underlying inequalities [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Married adolescent women aged 15\u0026ndash;19 years remain a high-risk and underserved group, demonstrating much lower contraceptive use than older women due to a combination of individual, relational, and contextual factors [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Therefore, in order to improve reproductive health outcomes and achieve Sustainable Development Goal 3.7, which is to ensure universal access to sexual and reproductive health services, it is vital to ensure that this population has equitable access to modern family planning methods [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eEpidemiological studies using survey data and logistic regression have established key determinants of contraceptive use in this group as parity, co-residency with a spouse, joint-decision making, exposure to the media, household income, and geographic location [\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. On the other hand, separation between spouses, particularly when that is caused by international migration of males, is a tremendous obstacle [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Geographically, such divisions as Mymensingh and Rajshahi are recording high usage rates as compared to Chittagong meaning that there are inequalities in terms of accessing services, social norms or program implementation [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Despite this progress, a critical gap remains. Traditional regression models, whilst establishing independent relationships, might not be able to reflect the non-linear, and higher-order interactions that are likely to define decision-making on contraception in reality [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. This limitation hinders the development of predictive models that are highly accurate and necessary for the precise targeting of interventions. Machine learning (ML) offers a powerful paradigm shift. Machine learning algorithms, including ensemble methods and neural networks, proficiently model complex patterns in high-dimensional data without rigid a priori assumptions regarding variable relationships, rendering them particularly effective for predictive tasks [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe application of machine learning to predict reproductive health is gaining popularity worldwide. Research conducted in Ethiopia and East Africa has effectively utilized algorithms such as Random Forest and XGBoost to predict contraceptive discontinuation and non-use, indicating their capability to surpass conventional approaches [\u003cspan additionalcitationids=\"CR22 CR23\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. In Bangladesh, initial research used machine learning to classify unintended pregnancy [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. However, there is a significant gap in the literature. To date, no study has systematically applied and compared multiple advanced machine learning algorithms to identify and interpret the key determinants of modern family planning use among married adolescent women. By combining feature selection, explainable machine learning techniques, and comparative model evaluation, the present study addresses this gap by identifying the most influential socio-demographic, relational, and contextual determinants of modern family planning use among married adolescent women in Bangladesh.\u003c/p\u003e \u003cp\u003eThis study uses data from the 2019\u0026ndash;20 Bangladesh Adolescent Health and Wellbeing Survey (BAHWS) to fill this gap. Specifically, the study aims to: (i) apply and systematically compare multiple machine learning models, including XGBoost, Random Forest, Neural Networks, and logistic regression, to identify the key determinants of modern family planning use among married adolescent women [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]; (ii) address class imbalance by using the Synthetic Minority Oversampling Technique (SMOTE) and assess the effect it has on model performance; and (iii) use sophisticated model interpretability techniques, especially SHapley Additive exPlanations (SHAP), to quantify and interpret the relative importance and directional effects of influential predictors [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. The findings are intended to inform targeted public health interventions and evidence-based policymaking aligned with Sustainable Development Goal targets, with the broader objective of improving reproductive health and child survival outcomes in Bangladesh.\u003c/p\u003e"},{"header":"Material and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData sources\u003c/h2\u003e \u003cp\u003eThe data for this study were derived from the Bangladesh Adolescent Health and Well-being Survey (BAHWS) 2019\u0026ndash;20, a nationwide cross-sectional survey that assesses the health and well-being of adolescents aged 15\u0026ndash;19 years in Bangladesh across a variety of factors. The BAHWS used a two-stage stratified sample strategy that included all administrative divisions and enumeration areas (EAs) throughout the country, assuring national representativeness. First, primary sampling units (PSUs) were identified, and subsequently, households were selected in a systematic manner during the second stage. PSUs were randomly classified into Type I and Type II, and questionnaires corresponding to PSU type (\u0026ldquo;Type One\u0026rdquo; or \u0026ldquo;Type Two\u0026rdquo;) were administered accordingly, following the official BAHWS sampling protocol [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe BAHWS collected information from ever-married adolescent women, never-married adolescent women, and never-married adolescent men aged 15\u0026ndash;19 years. The present analysis was restricted to ever-married adolescent women, as information on modern family planning use was collected for this subgroup. Of the 72,800 households selected for the survey, 97.7% responded, and interviews were successfully completed in 67,093 households. Among eligible ever-married adolescent women aged 15\u0026ndash;19 years, 4,926 respondents completed the interview, yielding a 97.2% response rate among 5,066 eligible individuals.\u003c/p\u003e \u003cp\u003eFor this study, we used the publicly accessible BAHWS 2019-20 dataset to extract records from ever-married adolescent women. Respondents who provided missing information on the outcome variable or key explanatory variables were excluded to ensure data quality. After these exclusions and survey weighting, the final analytic sample consisted of 3,223 ever-married teenage women aged 15 to 19, who were utilized for all descriptive analyses, machine learning model building, and regression validation.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eResponse variable\u003c/h3\u003e\n\u003cp\u003eThe outcome variable for this study was current use of modern family planning methods. In the BAHWS 2019\u0026ndash;20 survey, ever-married adolescent women were asked: \u0026ldquo;Are you or your partner currently doing something or using any method to delay or avoid getting pregnant?\u0026rdquo; Respondents who answered affirmatively were further asked to specify the method currently being used.\u003c/p\u003e \u003cp\u003eA binary outcome variable was created based on the answers to this question. Women were classified as modern family planning users (coded as 1) if they reported using any modern contraceptive method, such as female or male sterilization, intrauterine devices (IUD), injectables, implants, oral pills, male or female condoms, emergency contraceptive pills (ECP), or the lactational amenorrhea method (LAM). Women who reported not using any method or who used traditional methods, such as the rhythm (safe-period) method or withdrawal, were classified as non-users of modern family planning (coded as 0), in accordance with the World Health Organization (WHO) classification [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e].\u003c/p\u003e\n\u003ch3\u003eExplanatory variables\u003c/h3\u003e\n\u003cp\u003eGuided by prior literature [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan additionalcitationids=\"CR33\" citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e] and data availability in the BAHWS 2019\u0026ndash;20, the explanatory variables included age group (15\u0026ndash;17, 18\u0026ndash;19 years), educational attainment (0\u0026ndash;5, 6\u0026ndash;10, \u0026ge;\u0026thinsp;11 grades), employment status (yes/no), involvement in family planning programs in the past three years (yes/no), current residence with husband (living together, staying elsewhere in Bangladesh, staying outside Bangladesh), decision-making authority for family planning use (joint, mainly husband, mainly respondent, others), childbirth experience (ever given birth: yes/no), household wealth quintile (lowest to highest), exposure to mass media (frequency of watching television and reading newspapers/magazines), recent exposure to family planning messages through television or newspapers, place of residence (urban/rural), and administrative division.\u003c/p\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003eMachine learning modeling approach\u003c/h2\u003e \u003cp\u003eThis study used various supervised machine learning classification models to determine the factors that influence modern family planning method utilization among ever-married adolescent women. The outcome variable was binary, indicating the use of any modern contraceptive method (yes/no). The machine learning models used were Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree (DT), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), AdaBoost, Neural Network (Multi-Layer Perceptron), and Categorical Na\u0026iuml;ve Bayes (CNB). The analytic dataset was randomly split into training (70%) and testing (30%) subsets using stratified sampling to maintain the distribution of modern family planning users and non-users. To ensure reproducibility, a fixed random seed (random_state\u0026thinsp;=\u0026thinsp;42) was used.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eData preprocessing and encoding\u003c/h2\u003e \u003cp\u003eOne-hot encoding was used to transform categorical predictors in most models, whereas ordinal encoding was used in the Categorical Na\u0026iuml;ve Bayes model. Continuous variables were scaled using Min-Max normalization for models that are sensitive to feature scale, such as LR, SVM, KNN, and Neural Networks. Tree-based models were fitted with unscaled numerical features. This preprocessing was carried out using pipeline-based procedures to prevent data leaks during model training and evaluation.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eHandling class imbalance\u003c/h3\u003e\n\u003cp\u003eThe outcome variable exhibited a class imbalance, with a lower proportion of modern family planning non-users (27.34%\u003cb\u003e)\u003c/b\u003e than users (72.66%). To address this issue, most models' training data were treated with the Synthetic Minority Over-sampling Technique (SMOTE), resulting in a balanced 50:50 class distribution. For categorical features, SMOTEN was used in combination with the Categorical Na\u0026iuml;ve Bayes model. Model performance was assessed in both SMOTE-adjusted and non-SMOTE conditions to evaluate robustness.\u003c/p\u003e\n\u003ch3\u003eHyperparameter optimization\u003c/h3\u003e\n\u003cp\u003eWe used GridSearchCV with five-fold cross-validation on the training dataset to find the best hyperparameters. Various performance metrics were evaluated during the tuning process. This method guaranteed optimal model calibration for identifying modern family planning users while addressing class imbalance.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eModel evaluation and validation\u003c/h2\u003e \u003cp\u003eModel performance on the held-out test dataset was assessed using accuracy, precision, recall (sensitivity), F1-score, Matthew\u0026rsquo;s correlation coefficient (MCC), Cohen's Kappa, and AUROC. Confusion matrices were created to evaluate classification errors in detail. Additional k-fold cross-validation was performed with different fold sizes (5, 10, 15, 20, 25, and 30 folds) in order to ensure robustness and generalizability.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eReceiver operating characteristic analysis\u003c/h2\u003e \u003cp\u003eThe ROC curve graphically represents the trade-off between true positives and false positives at the various classification thresholds [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. ROC curves were generated for all fitted models to illustrate the discrimination performance across classification thresholds. The performance of models was compared using the AUROC, with higher values indicating better discrimination between modern family planning users and non-users.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eFeature selection and explainability\u003c/h2\u003e \u003cp\u003eThe Boruta algorithm was used to select the most important predictors of modern family planning use. Boruta compares the relevance of observed variables to randomized shadow features in a Random Forest framework, retaining only statistically significant predictors [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. To improve model interpretability, feature importance from tree-based models and SHapley Additive Explanations (SHAP) were used to quantify the direction and magnitude of each predictor\u0026rsquo;s contribution to model predictions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eComparison with traditional statistical methods\u003c/h2\u003e \u003cp\u003eIn addition to machine learning methods, weighted bivariate analyses with chi-square tests were performed in R to investigate the relationships between modern family planning use and explanatory variables. A multivariable logistic regression model was also fitted as a validation step to assess consistency between statistically significant predictors and those identified by machine learning models.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Software and Implementation\u003c/h2\u003e \u003cp\u003eAll machine learning studies were performed in Python (version 3.11), while R (version 4.4.3) was used for bivariate analysis, feature selection with the Boruta algorithm, geographic visualizations, and survey-weighted logistic regression. Using both software platforms enabled us to leverage Python\u0026rsquo;s strengths in machine learning and R\u0026rsquo;s capabilities in statistical analysis and feature selection, resulting in a robust, reproducible analytical framework.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eThe socio-demographic and reproductive characteristics of married adolescent women aged 15\u0026ndash;19 years, as well as their association with modern family planning (FP) use, are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Overall, 72.66% (n\u0026thinsp;=\u0026thinsp;2,342) of respondents reported utilizing a modern FP method, whereas 27.34% (n\u0026thinsp;=\u0026thinsp;881) did not. The majority (57.1%) were between the ages of 18 and 19, while 75.7% had completed grades 6 through 10, and 93.8% were unemployed. Most participants had not been involved in FP programs during the previous three years (95.0%). Regarding marriage and having children, 85.3% were living with their husbands, 82.5% said they made FP decisions together, and 54.8% had given birth at some point. Media exposure was also different: almost half (49.4%) watched television daily, and 86.9% did not read newspapers or magazines. Most of them (77.1%) lived in rural areas, and the highest percentage of respondents (24.3%) was from the Dhaka division.\u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, chi-square tests revealed significant differences in modern FP use across several variables. FP use was higher among adolescents living with their husbands (77%), those sharing FP decisions with their partners (74.1%), and those who had ever given birth (80.7%) (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\:\\)\u003c/span\u003e\u003c/span\u003e\u0026lt; 0.001 for all). Exposure to mass media was also associated with FP use, women who watched television daily (74.5%) or read newspapers at least once a week (82.6%) showed higher use compared to those with no exposure (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\:\\)\u003c/span\u003e\u003c/span\u003e\u0026lt; 0.05).\u003c/p\u003e \u003cp\u003eThe regional difference was also significant (\u0026lt;\u0026thinsp;0.001), with the highest rates of modern FP usage in Mymensingh (80.1%) and Rajshahi (79%), followed by Rangpur (75.7%) and Barisal (73.9%), with the lowest in Chittagong (63.5%). Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows a choropleth map of regional disparities in modern family planning (FP) use across Bangladesh, with darker shades indicating divisions with a higher prevalence of modern FP use. These spatial patterns highlight regional disparities in FP uptake, whereas no statistically significant differences were observed at the 5% level by age, education, employment, FP program participation, wealth, residence, or FP messages on television.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBackground Characteristics of Married Adolescent Women and Bivariate Associations with Modern Family Planning Method Use, BAHWS 19\u0026ndash;20.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eCharacteristics\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{N}\\)\u003c/span\u003e\u003c/span\u003e (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c5\" namest=\"c3\"\u003e \u003cp\u003eModern Family Planning Method Use\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003eYes\u003c/b\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{N}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e(%)\u003c/b\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eNo\u003c/b\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{N}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e(%)\u003c/b\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{p}\\)\u003c/span\u003e\u003c/span\u003e \u003cb\u003e-value\u003c/b\u003e\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAge group\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e15\u0026ndash;17\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1382 (42.87)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1016 (73.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e365 (26.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.347\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e18\u0026ndash;19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1841 (57.13)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1326 (72%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e516 (28%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEducational grades\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e0\u0026ndash;5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e601 (18.64)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e456 (75.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e145 (24.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.066\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e6\u0026ndash;10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2438 (75.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1761 (72.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e677 (27.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e11 and Higher\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e184 (5.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e125 (67.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e59 (32.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEmployment status\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3024 (93.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2204 (72.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e820 (27.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.294\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e199 (6.18)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e138 (69.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e61 (30.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eInvolved in FP program in last 3 years\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3062 (95.01)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2219 (72.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e842 (27.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.392\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e161 (4.99)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e122 (75.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e39 (24.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCurrently residing with husband\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLiving with her\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2750 (85.34)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2117 (77%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e633 (23%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStaying elsewhere in Bangladesh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e299 (9.27)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e204 (68.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e95 (31.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eStaying elsewhere outside Bangladesh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e174 (5.39)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21 (12.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e153 (87.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDecision maker for FP use\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBoth\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2658 (82.48)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1971 (74.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e688 (25.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMainly husband\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e275 (8.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e166 (60.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e109 (39.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMainly my decision\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e259 (8.02)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e187 (72.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e72 (27.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e31 (0.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18 (58%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e13 (42%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eEver given birth\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1457 (45.22)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e916 (62.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e541 (37.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1766 (54.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1426 (80.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e340 (19.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eWealth quantile\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLowest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e611 (18.95)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e449 (73.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e161 (26.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"4\" rowspan=\"5\"\u003e \u003cp\u003e0.064\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSecond\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e711 (22.05)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e541 (76.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e170 (23.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMiddle\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e725 (22.50)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e517 (71.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e208 (28.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRicher\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e686 (21.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e497 (72.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e190 (27.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHighest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e490 (15.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e337 (68.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e153 (31.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eWatch TV\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAt least once a week\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e526 (16.31)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e390 (74.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e136 (25.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003e0.019\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEvery day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1591 (49.36)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1185 (74.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e406 (25.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLess than one a week\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e256 (7.95)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e174 (67.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e82 (32.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNot at all\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e850 (26.38)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e593 (69.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e257 (30.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRead Newspaper/Magazine\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAt least once a week\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e128 (3.97)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e101 (78.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e27 (21.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"3\" rowspan=\"4\"\u003e \u003cp\u003e0.008\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEvery day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e20 (0.62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e16 (82.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3 (17.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLess than one a week\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e276 (8.57)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e179 (65%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e97 (35%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNot at all\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2799 (86.85)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2045 (73.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e754 (26.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHeard about FP on TV (Last Few Months)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2767 (85.86)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1999 (72.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e768 (27.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.197\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e456 (14.14)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e343 (75.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e113 (24.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c3\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRead about FP in News (Last Few Months)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3179 (98.64)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2303 (72.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e876 (27.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003e0.040\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e44 (1.36)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e38 (87.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 (12.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eResidence\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther urban\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e150 (4.67)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e105 (69.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e46 (30.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003e0.480\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRural\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2484 (77.06)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1800 (72.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e684 (27.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUrban\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e589 (18.28)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e437 (74.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e152 (25.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAdministrative Division\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBarisal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e214 (6.64)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e158 (73.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e56 (26.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\" morerows=\"7\" rowspan=\"8\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChittagong\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e406 (12.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e257 (63.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e148 (36.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDhaka\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e783 (24.30)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e548 (70%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e235 (30%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKhulna\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e438 (13.57)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e303 (69.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e135 (30.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMymensingh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e304 (9.44)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e244 (80.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e61 (19.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRajshahi\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e569 (17.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e450 (79%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e119 (21%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRangpur\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e437 (13.57)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e331 (75.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e106 (24.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSylhet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e72 (2.24)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e51 (70.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e21 (29.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTotal\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e3223 (100.0)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e2342 (72.66)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e881 (27.34)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eFeature Selection\u003c/h2\u003e \u003cp\u003eThe Boruta algorithm was used to determine the most important features of modern contraceptive use among the original set of variables by iteratively comparing the feature importance scores with those of randomly permuted shadow features. This procedure approved features whose importance was greater than the maximum shadow value as significant and discarded those whose importance was less than the minimum or average shadow values. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, the Boruta analysis identified a set of key predictors that are significantly associated with modern FP use. The most influential features were currently residing with the husband, having ever given birth, age group, decision-making authority for FP use, and administrative division. Other variables, such as media exposure (TV and newspapers), wealth quantile, and participation in FP programs, also contributed moderately to model performance. Conversely, education, employment status, and place of residence had a small predictive power and were considered to be less important in further training of the model.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eClass Balancing Using SMOTE\u003c/h2\u003e \u003cp\u003eTo balance the outcome variable, we used the Synthetic Minority Oversampling Technique (SMOTE) that creates synthetic observations in the minority class to balance the imbalance between the classes. Before balancing, the dataset had an unequal distribution, with 2,290 (72.34%) adolescents utilizing modern family planning (FP) methods and 876 (27.66%) not. After applying SMOTE, an equal number of observations were created for both categories, resulting in a symmetrical 50:50 distribution of modern FP users and non-users.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eModel Performance Comparison for predicting the Modern FP method use\u003c/h2\u003e \u003cp\u003eSeveral machine learning algorithms, such as Random Forest (RF), Decision Tree (DT), K-Nearest Neighbors (KNN), Logistic Regression (LR), Support Vector Machine (SVM), XGBoost, LightGBM, AdaBoost, Categorical Na\u0026iuml;ve Bayes, and Neural Networks, were used to develop a predictive model for modern FP method use among married adolescents in Bangladesh. Each model was trained using 70% of the data, with the remaining 30% for testing. Hyperparameter tuning was performed via GridSearchCV with 5-fold cross-validation to optimize model parameters which are provided in Supplementary Table\u0026nbsp;1. Model performance was evaluated using accuracy, precision, recall, F1-score, AUC (Area Under the Curve), Matthews Correlation Coefficient (MCC), and Cohen\u0026rsquo;s Kappa (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). To address class imbalance, the evaluation included both imbalanced (original data) and balanced (SMOTE) scenarios, with XGBoost, Neural network, RF, and LR emerging as top performers. Models trained on the unbalanced data set achieved high overall accuracy but were evidently biased toward the majority group (FP users). The XGBoost and Random Forest models exhibited the highest accuracies (0.7811 each), accurately predicting approximately 78% of all cases. In the case of XGBoost, precision (0.7782), recall (0.9753), and F1-score (0.8656) were high and AUC of 0.7559 indicated good separation of classes, but the classification report indicated low recall of non-use (class 0: precision 0.81, recall 0.27, F1-score 0.41), reflecting poor performance of minority classes. MCC (0.3824) and Kappa (0.3129) of Cohen showed moderate reliability. RF also exhibited a similar pattern (accuracy\u0026thinsp;=\u0026thinsp;0.7811; AUC\u0026thinsp;=\u0026thinsp;0.7550), with very high recall among FP users (0.9884) but lower recognition among non-users.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance Metrics of Machine Learning Models for Modern FP Use Prediction with 95% Confidence Intervals, BAHWS 19\u0026ndash;20.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eML Models\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePeformance\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eUnbalanced (Without SMOTE)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eBalanced (With SMOTE)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eLogistic Regression\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7716\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0274\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7147\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0295\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7537\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0334\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.7503\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0330\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3519\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0610\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3597\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0600\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.2512\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0595\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3522\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0600\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7640\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0282\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8455\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0282\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.9898\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0075\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7409\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0345\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.8624\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0182\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7898\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0250\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eSVM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7747\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0268\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.6611\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0305\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.6791\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0375\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7285\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0343\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3704\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0561\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3271\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0585\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.2558\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0584\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3011\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0563\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7642\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0281\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8600\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0278\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.9956\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0050\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.6346\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0363\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.8647\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0186\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7303\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0280\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eKNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7611\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0268\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7274\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0279\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7242\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0354\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7067\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0377\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3030\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0647\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.2992\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0679\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.2214\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0571\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.2985\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0682\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7596\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0279\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8006\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0295\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.9796\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0095\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8297\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0277\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.8557\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0185\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8149\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0215\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eNeural Network\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7716\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0263\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7358\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0279\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7509\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0347\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7453\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0353\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3469\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0648\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3708\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0605\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.2670\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0609\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3690\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0611\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7683\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0270\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8364\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0272\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.9796\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0108\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7889\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0310\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.8612\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0181\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8120\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0219\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eRandom Forest\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.7811\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0268\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7337\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0274\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7550\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0350\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7447\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0356\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.3878\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0608\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3435\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0642\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.2942\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0623\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3434\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0643\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7725\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0280\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8210\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0266\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.9884\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0080\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8079\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0286\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.8672\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0184\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8144\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0218\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eDecision Tree\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7737\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0269\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7126\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0279\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7283\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0338\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.6953\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0404\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3568\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0601\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.2983\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0651\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.2693\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0592\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.2979\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0653\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7682\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0276\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8099\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0283\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.9840\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0089\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7875\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0296\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.8628\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0183\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7985\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0223\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eXGBoost\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7811\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0263\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.7600\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0263\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7559\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0340\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7285\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0379\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3824\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0655\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.3720\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0663\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.3129\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0639\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.3695\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0660\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7782\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0279\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8148\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0269\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.9753\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0115\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.8646\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0254\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.8656\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0182\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.8390\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0201\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eLightGBM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7642\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0268\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7337\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0279\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7120\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0384\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.6861\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0416\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3316\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0722\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3063\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0678\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3013\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0680\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3047\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0680\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.7820\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0277\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7997\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0285\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.9345\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0188\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8428\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0259\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.8515\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0192\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8207\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0213\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eAdaBoost\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7726\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0258\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7253\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0274\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.7581\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0340\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7350\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0349\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3500\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0653\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3488\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0604\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.2822\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0635\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.3468\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0604\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7723\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0286\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8307\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0280\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.9723\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0120\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7787\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0318\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.8608\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8039\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0232\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"6\" rowspan=\"7\"\u003e \u003cp\u003e\u003cb\u003eCategorical Na\u0026iuml;ve Bayes\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAccuracy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7716\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0268\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.6926\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0284\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7377\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0343\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7235\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0362\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eMCC\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.3460\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0654\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.2895\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0647\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eCohen Kappa\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.2757\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0613\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.2860\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0638\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003ePrecision\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.7707\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0273\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.8170\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0296\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eRecall\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.9738\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0122\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7409\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0317\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eF1-score\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.8605\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.7771\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0242\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eLightGBM showed slightly lower accuracy (0.7642) but maintained a reasonable AUC (0.7120) and F1-score (0.8515). The Neural Network (AUC\u0026thinsp;=\u0026thinsp;0.7509) and Decision Tree (AUC\u0026thinsp;=\u0026thinsp;0.7283) gave moderate results. Other less sophisticated models, such as LR (AUC\u0026thinsp;=\u0026thinsp;0.7537), SVM (AUC\u0026thinsp;=\u0026thinsp;0.6791), and KNN (AUC\u0026thinsp;=\u0026thinsp;0.7242), did not perform well, with lower discriminative power and weaker agreement (e.g., MCC\u0026thinsp;=\u0026thinsp;0.3519 for LR). Overall, models performed well on dominant patterns but struggled to capture minority (non-user) traits.\u003c/p\u003e \u003cp\u003eBalancing the dataset with SMOTE increased sensitivity to the minority class, resulting in a more symmetric distribution of predictions. XGBoost performed best overall (accuracy\u0026thinsp;=\u0026thinsp;0.7600; AUC\u0026thinsp;=\u0026thinsp;0.7285), with high discrimination and stable agreement (MCC\u0026thinsp;=\u0026thinsp;0.3720; κ\u0026thinsp;=\u0026thinsp;0.3695). Its classification report demonstrated improved performance for non-users (precision 0.58, recall 0.49, F1 0.53) and consistent accuracy for users (precision 0.81, recall 0.86, F1 0.84), with an overall macro-F1 of 0.68 and weighted-F1 of 0.75. The Random Forest model also performed well (accuracy\u0026thinsp;=\u0026thinsp;0.7337; AUC\u0026thinsp;=\u0026thinsp;0.7447; MCC\u0026thinsp;=\u0026thinsp;0.3435; k\u0026thinsp;=\u0026thinsp;0.3434), with a high recall (0.8079) and F1-score (0.8144). The neural network achieved similar stability (accuracy\u0026thinsp;=\u0026thinsp;0.7358; AUC\u0026thinsp;=\u0026thinsp;0.7453; MCC\u0026thinsp;=\u0026thinsp;0.3708), indicating reliable generalization under balanced conditions. AdaBoost and LightGBM demonstrated competitive performance (AUCs\u0026thinsp;\u0026asymp;\u0026thinsp;0.69\u0026ndash;0.74), while DT (AUC\u0026thinsp;=\u0026thinsp;0.6953) and KNN (AUC\u0026thinsp;=\u0026thinsp;0.7067) demonstrated moderate improvements. Simpler classifiers, LR (AUC\u0026thinsp;=\u0026thinsp;0.7503) and SVM (AUC\u0026thinsp;=\u0026thinsp;0.7285), had slightly higher Cohen\u0026rsquo;s Kappa values (0.3522 for LR), but their accuracies (0.7147 and 0.6611) were lower than ensemble counterparts. The Categorical Na\u0026iuml;ve Bayes model showed modest performance (AUC\u0026thinsp;=\u0026thinsp;0.7235). Collectively, Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e shows that ensemble-based and deep learning models performed significantly better than standard classifiers. XGBoost received the highest overall composite score across all seven metrics, followed by Neural Network, Random Forest, and AdaBoost. These models consistently delivered greater recall, AUC, and MCC values, essential for imbalanced contexts, while retaining balanced accuracy. LightGBM and LR models provided adequate performance but slightly less stability. In contrast, DT, KNN, Na\u0026iuml;ve Bayes, and SVM were the weakest performers, characterized by lower Kappa (0.29\u0026ndash;0.33) and MCC (\u0026lt;\u0026thinsp;0.31).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows the ROC curves for ten machine learning models evaluated on the SMOTE balanced test dataset. The ROC curve illustrates each model's ability to distinguish between users and non-users of modern family planning (FP) methods among married adolescent women. Among all models, Logistic Regression (AUC\u0026thinsp;=\u0026thinsp;0.7503), Neural Network (AUC\u0026thinsp;=\u0026thinsp;0.7453), and Random Forest (AUC\u0026thinsp;=\u0026thinsp;0.7447) showed the most effective separation between the two groups, demonstrating a reasonable balance between sensitivity and specificity. AdaBoost (AUC\u0026thinsp;=\u0026thinsp;0.7350), XGBoost and SVM (AUC\u0026thinsp;=\u0026thinsp;0.7285) performed well, whereas Categorical Na\u0026iuml;ve Bayes provided moderate discrimination (AUC\u0026thinsp;=\u0026thinsp;0.7235). KNN (AUC\u0026thinsp;=\u0026thinsp;0.7067), Decision Tree (AUC\u0026thinsp;=\u0026thinsp;0.6953), and LightGBM (AUC\u0026thinsp;=\u0026thinsp;0.6861) performed less effectively, reflecting slightly weaker generalization to unseen data. Overall, ensemble and neural network models provided the most reliable classification results, while Logistic Regression, while being a simpler linear model, achieved one of the highest AUCs, demonstrating its strong interpretability and consistent predictive power.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eK-fold cross validation\u003c/h2\u003e \u003cp\u003eTo ensure model stability and generalizability, a K-fold cross-validation process of 5, 10, 15, 20, 25, and 30 folds was used. Mean Accuracy (MAcc) and Precision\u0026ndash;Recall AUC (PR-AUC) were calculated for each fold configuration across all models (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAcross all folds, XGBoost, K-Nearest Neighbors (KNN), and Random Forest (RF) outperformed the other models. XGBoost maintained a high and consistent accuracy of 0.72\u0026ndash;0.73, with a PR-AUC of 0.84\u0026ndash;0.85, demonstrating strong discrimination between users and non-users of modern family planning methods. KNN showed high precision and stability across folds, with PR-AUC ranging from 0.82 to 0.83 and MAcc ranging from 0.71 to 0.72. With an MAcc of roughly 0.71 and a PR-AUC of 0.84\u0026ndash;0.85, Random Forest also yielded consistent results, demonstrating robust, generalizable performance across fold sizes. Logistic Regression and Neural Network produced consistent results, with MAcc averaging 0.68 and PR-AUC ranging from 0.84 to 0.85 across all folds. AdaBoost demonstrated comparable stability, with MAcc around 0.70 and PR-AUC ranging from 0.84 to 0.85, indicating balanced and consistent predictive performance. LightGBM also performed well, with MAcc ranging between 0.70 and 0.71 and PR-AUC around 0.82, indicating efficient generalization to unseen data. Conversely, Decision Tree and Categorical Na\u0026iuml;ve Bayes had marginally inferior and less consistent performance, with Decision Tree achieving MAcc between 0.68 and 0.69 and PR-AUC between 0.80 and 0.82, whereas Na\u0026iuml;ve Bayes attained MAcc between 0.68 and 0.69 and PR-AUC between 0.83 and 0.84. The SVM showed the lowest accuracy, with a mean accuracy (MAcc) ranging from 0.63 to 0.64, while sustaining a PR-AUC of approximately 0.84, reflecting limited improvement across folds. Overall, the findings indicate that XGBoost, Random Forest and KNN are the most effective and stable models across all fold configurations.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eK-fold cross-validation performance (Mean Accuracy and PR-AUC) of the selected machine learning models.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"13\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c13\" colnum=\"13\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e5 Folds\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003e10 Folds\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e \u003cp\u003e15 Folds\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003e20 Folds\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c11\" namest=\"c10\"\u003e \u003cp\u003e25 Folds\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c13\" namest=\"c12\"\u003e \u003cp\u003e30 Folds\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMAcc\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePR-AUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eMAcc\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePR-AUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eMAcc\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePR-AUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eMAcc\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003ePR-AUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eMAcc\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003ePR-AUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c12\"\u003e \u003cp\u003eMAcc\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c13\"\u003e \u003cp\u003ePR-AUC\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLogistic Regression\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSVM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.63\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.64\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eKNN\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.72\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eNeural Network\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.84\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRandom Forest\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDecision Tree\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eXGBoost\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e0.73\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.84\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.72\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e\u003cb\u003e0.84\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e\u003cb\u003e0.72\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e\u003cb\u003e0.84\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e\u003cb\u003e0.72\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e\u003cb\u003e0.72\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e\u003cb\u003e0.85\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLightGBM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAdaBoost\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.70\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCategorical Na\u0026iuml;ve Bayes\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.69\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c12\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c13\"\u003e \u003cp\u003e0.84\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eFeature Selection Results and Model Explainability\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e displays the feature importance rankings obtained from the XGBoost model, highlighting the principal factors affecting modern family planning (FP) utilization among married adolescent women in Bangladesh. The variable \u0026ldquo;Currently residing with husband: Staying elsewhere outside Bangladesh\u0026rdquo; was the most significant predictor (importance\u0026thinsp;=\u0026thinsp;0.218), indicating that women whose spouses reside overseas are considerably less likely to use modern family planning methods. The second most significant variable was \u0026ldquo;Ever given birth: No\u0026rdquo; (0.065), indicating that women who have not yet had childbirth often exhibit lower family planning utilization. Regional variation also played a substantial role, with Divisions such as Chittagong, Mymensingh, Rajshahi, and Dhaka exhibiting significant notable scores (ranging\u0026thinsp;=\u0026thinsp;0.040\u0026ndash;0.046), indicating persistent geographical disparities in FP access and uptake. Additionally, indicators of media exposure, particularly the frequency of reading newspapers or magazines weekly, were significant. The model\u0026rsquo;s predictions were also influenced by economic status variables, such as wealth quintiles (from lowest to wealthiest), and program-related factors (e.g., involvement in FP programs in the last 3 years and decision-making authority for FP use). These findings highlight the importance of both structural and relational factors in understanding variations in modern FP use among adolescent women.\u003c/p\u003e \u003cp\u003eThe integrated SHAP summary and feature importance plot (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e) shows the 20 most significant features identified by the XGBoost model for predicting modern family planning (FP) use. The features are ranked vertically by their mean absolute SHAP values, which indicate their average contribution to model predictions. Each dot represents an individual observation and is color-coded by feature value (red=high, blue\u0026thinsp;=\u0026thinsp;low), indicating how increases or decreases in that variable influence the probability of using modern FP. The strongest predictor was \u0026ldquo;Ever given birth: No\u0026rdquo; (mean |SHAP| = 0.449), with smaller SHAP values (blue) consistently reducing the predicted probability of modern FP use. This finding reflects that adolescent women who have not yet had a child are substantially less likely to adopt modern contraceptives. Conversely, women who have given birth (high feature values in red) were more likely to use FP methods. The second significant feature, \"Currently residing with husband: Staying elsewhere outside Bangladesh\" (mean |SHAP| = 0.294), had mostly negative SHAP values, indicating a lower likelihood of FP use when the husband resides abroad. The third related variable, \"Currently residing with husband: Living with her,\" exhibited the opposite effect, contributing positively to the likelihood of FP use.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eRegional variables, including Chittagong, Rajshahi, Mymensingh, Khulna, and Dhaka divisions (0.034\u0026ndash;0.105), exhibited mixed SHAP effects, reflecting both positive and negative influences depending on location. Notably, adolescents in Chittagong and Mymensingh divisions had higher SHAP variability, implying greater regional inequality in FP access and awareness.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eSocioeconomic status indicators, especially wealth quintiles (lowest to highest), exhibited a clear gradient. Wealth quintiles showed a clear directional gradient: higher wealth (red) was positively associated with modern FP use, while lower wealth (blue) was negatively associated, indicating a reduced probability of modern FP use. Media exposure variables, such as \u0026ldquo;Read any newspaper/magazine at least once a week,\u0026rdquo; had a positive impact on SHAP, whereas \u0026ldquo;Read not at all\u0026rdquo; had a negative impact. Behavioral and decision-related features, such as \u0026ldquo;Decision maker for FP use: Both\u0026rdquo; and \u0026ldquo;Involved in FP program in last 3 years: Yes,\u0026rdquo; exhibited positive SHAP values, suggesting increased probabilities of FP use. Conversely, their counterparts (\u0026ldquo;No\u0026rdquo; or \u0026ldquo;Mainly husband\u0026rdquo;) exhibited negative contributions. Overall, high SHAP values for childbirth history, shared FP decisions, and wealthier households indicate a greater likelihood of FP use, while negative SHAP values for spousal absence and poverty indicate reduced FP use.\u003c/p\u003e \u003cp\u003eThe fitted LR model in Supplementary Table\u0026nbsp;2 revealed that several explanatory variables remained statistically significant (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e \u0026lt; 0.05), aligning closely with the key predictors identified through SHAP analysis. Adolescents aged 18\u0026ndash;19 years were significantly less likely to use modern FP methods compared to those aged 15\u0026ndash;17 years (AOR\u0026thinsp;=\u0026thinsp;0.71; 95% CI: 0.58\u0026ndash;0.85; \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e = 0.0003), which aligns with the SHAP analysis results, also showing a positive value for the 15\u0026ndash;17 age group. Participation in FP decision-making as \u0026ldquo;Both partners\u0026rdquo; (AOR = 1.61; 95% CI: 1.20\u0026ndash;2.16; \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e = 0.0015) was positively associated with FP use, consistent with SHAP findings. Women who had ever given birth showed substantially higher odds of FP use (AOR = 3.20; 95% CI: 2.62\u0026ndash;3.89; \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e \u0026lt; 0.0001), reaffirming its dominant importance from the SHAP model. Those whose husbands were staying elsewhere in Bangladesh (AOR = 0.74, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e = 0.044) or outside the country (AOR = 0.03, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e \u0026lt; 0.001) had lower odds, matching SHAP\u0026rsquo;s strong negative effects for spousal absence. Reading newspapers at least once a week (AOR = 1.72, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e = 0.028) and joint decision-making with the husband (AOR = 1.61,\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\:p\\)\u003c/span\u003e\u003c/span\u003e = 0.002) were associated with higher FP use, consistent with positive SHAP contributions. Similarly, wealthier households (AOR range = 1.38\u0026ndash;1.65, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e \u0026lt; 0.05) and residence in Mymensingh (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e = 0.027) and Rajshahi (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\)\u003c/span\u003e\u003c/span\u003e = 0.023) divisions showed positive effects, paralleling SHAP findings of regional and socioeconomic influence.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe modern family planning (FP) methods are necessary to empower married adolescent women so that they can have control over their reproductive health, especially in Bangladesh, where early marriage is still a common practice and causes high levels of unwanted pregnancies and maternal deaths [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. This study used a comprehensive machine learning (ML) framework to predict modern family planning (FP) use among married adolescent women aged 15\u0026ndash;19 years in Bangladesh using nationally representative BAHWS 2019-20 data. The prevalence of modern FP use was 72.66%, with significant geographical differences: Mymensingh and Rajshahi had the highest rates, while Chittagong had the lowest. Ensemble methods were more effective in prediction across models, particularly when SMOTE was used to mitigate class imbalance. XGBoost turned out to be the best-performing model, with improved accuracy, AUC, precision, and recall compared to other models. The most important determinants were spousal co-residency (husband living abroad and husband living with her), parity (ever having a baby), administrative division, FP decision-making dynamics, age group, media exposure, and wealth quantile, with strong consistency between SHAP analysis and survey-weighted logistic regression.\u003c/p\u003e \u003cp\u003eThe prevalence of modern FP use in our adolescent sample exceeds the national adolescent contraceptive prevalence rate from the BDHS summary findings [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], possibly due to our focus on married current users and survey-specific sampling. However, it remains lower than among older women, highlighting disparities driven by adolescents\u0026rsquo; limited autonomy and relational barriers [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. Regional clustering, visualized in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, aligns with BDHS analyses showing southeastern divisions like Chittagong facing service inequities and conservative norms [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. The finding is higher than in high-fertility Sub-Saharan African countries[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] but lower than in urban Kenya [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e], reflecting Bangladesh\u0026rsquo;s progress in FP programs amid persistent adolescent gaps.\u003c/p\u003e \u003cp\u003eTen ML algorithms were trained on balanced and unbalanced datasets, with performance evaluated using accuracy, AUC, precision, recall, F1-score, Matthews Correlation Coefficient (MCC), and Cohen\u0026rsquo;s Kappa. In the unbalanced dataset, several models achieved relatively high overall accuracy but showed poor sensitivity to the minority class (non-users), indicating majority-class bias. This is a common issue in health prediction problems where the outcome distribution is asymmetric, and it highlights why relying only on accuracy can be misleading [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. After applying SMOTE, performance became more balanced, and ensemble models remained superior. XGBoost achieved the best overall composite performance across the evaluated metrics and also showed strong stability in k-fold cross-validation, a finding which aligns with a similar study on modern FP use among reproductive-age women in Ethiopia, where the XGBoost model outperformed other classifiers [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. Our findings are consistent with evidence from other ML studies, in which XGBoost and Random Forest often outperform conventional classifiers in predicting FP-related outcomes such as modern FP non-use, optimal ANC utilization, or fertility preferences [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. SMOTE addressed class imbalance by improving recall among the minority class (non-users), which is consistent with improvements in Ethiopian fertility models [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. This robustness, with ensemble methods outperforming simpler classifiers such as Decision Tree (DT) or K-Nearest Neighbors (KNN), demonstrates machine learning's shift from linear assumptions in chi-square tests to capturing complex interactions [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Unlike logistic regression, which may fail due to multicollinearity and other statistical violations, ensemble methods effectively capture the complex, nonlinear interactions found in public health data, making them more reliable for prediction in this study [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSpousal co-residency emerged as the most influential determinant across Boruta selection, XGBoost importance, and SHAP analyses. Adolescents whose husbands lived outside Bangladesh had extremely low FP use, which is consistent with studies in Bangladesh [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e] and India [\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e] that found reduced contraceptive uptake among women experiencing spousal separation due to migration. In contrast, research conducted in East Africa that employs machine learning methods identifies marital status (married vs. unmarried) as a significant predictor, rather than spousal co-residency [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]. This discrepancy is likely due to contextual differences in Bangladesh, where nearly all adolescents in the sample were married; spousal co-residency serves as a more informative marker of FP use than marital status alone. Adolescents residing with their husbands were significantly more likely to use modern family planning, a finding that was consistently prioritized across Boruta selection, XGBoost importance, and SHAP analyses. This is consistent with patterns in South Asia, where co-residency with husbands and, in some cases, with other family members such as mothers-in-law, is associated with higher use of modern contraceptive methods [\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]. Adolescents who had ever given birth were substantially more likely to use modern FP, consistent with previous Bangladesh studies showing that contraceptive adoption often follows the first birth rather than preceding it [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eJoint decision-making for FP use showed a positive association with FP uptake, aligning with prior findings from Bangladesh and South Asia that emphasize the role of couple communication and shared authority [\u003cspan additionalcitationids=\"CR53\" citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e]. This result is consistent with an ML-based FP study in East Africa that highlights women\u0026rsquo;s autonomy and decision-making power as key predictors, even though it modeled the non-use of family planning, unlike this study\u0026rsquo;s focus on FP uptake [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Socioeconomic status also played an important role in modern FP use. Adolescents from wealthier households were more likely to use modern FP methods, while those in the lowest wealth quintile had lower uptake. This finding is consistent with previous studies in Bangladesh and Nigeria that demonstrate that economic disadvantage limits access to contraceptive services and information, thus reducing utilization [\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e, \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e]. Although the bivariate association between wealth and FP use was marginal in this study, the SHAP results suggest that wealth interacts with other factors, such as geographic location and media exposure, in complex ways that may not be fully captured by traditional statistical tests.\u003c/p\u003e \u003cp\u003eExposure to media, particularly reading newspapers or magazines, was positively associated with the use of modern FP. Adolescents with no exposure to newspapers or magazines were less likely to use FP, whereas those with regular exposure had higher uptake. This aligns with prior evidence in Ethiopia, indicating that access to information through TV, radio, and magazines increases awareness of modern contraceptive options and improves informed decision-making [\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e]. Age differences were also observed in modern FP use among married adolescents. Although the chi-square test indicated no significant bivariate association between age group and modern FP use, Boruta feature selection and SHAP analysis identified age group as an important predictor, with the 15\u0026ndash;17-year age group contributing positively to the predicted FP probability compared to 18\u0026ndash;19-year-olds. Adolescents aged 15\u0026ndash;17 years were more likely to use modern FP methods than those aged 18\u0026ndash;19 years, after adjustment for other factors. This aligns with a systematic review in Bangladesh that found younger women utilize contraceptives more frequently than older women [\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e]. It also corresponds with DHS evidence from 2017\u0026ndash;18, which indicates that younger contraceptive users (ages 15\u0026ndash;24) are more likely to obtain methods from the private sector than older users [\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eGeographic variation in FP use remained pronounced in this study. Adolescents residing in Mymensingh and Rajshahi divisions had higher modern FP use than those in Chittagong, which consistently showed lower uptake. These findings are consistent with prior Bangladesh research showing persistent geographic inequalities in FP uptake across divisions, reflecting differences in service delivery, program intensity, infrastructure, and sociocultural norms [\u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e, \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e61\u003c/span\u003e]. Participation in FP programs over the past three years had a modest effect on FP use. While program involvement contributed positively in the SHAP analyses, it did not emerge as a strong independent predictor in conventional regression models.\u003c/p\u003e \u003cp\u003eThe integration of SHAP and Boruta allowed this study to move beyond prediction to interpretation. Similar to a recent ML study on modern FP outcome [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e] SHAP revealed directional effects and interaction patterns not easily captured by traditional regression. The close alignment between SHAP-derived important predictors and survey-weighted logistic regression results supports the validity of the ML framework.\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section2\"\u003e \u003ch2\u003eStrengths, limitations, and future directions\u003c/h2\u003e \u003cp\u003eThis study\u0026rsquo;s strengths include the use of nationally representative adolescent data, applying major data preprocessing, systematic comparison of multiple ML models, explicit handling of class imbalance, and integration of SHAP analysis with regression validation. Limitations include the cross-sectional design, reliance on self-reported FP use, exclusion of unmarried adolescents, limited data on proximal determinants, and lack of external validation. Future research should incorporate longitudinal data, include proximal factors like autonomy and service quality, and apply deep learning.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study applied and compared multiple machine learning classification models to predict modern family planning (FP) use and identify its key determinants among married adolescent women aged 15\u0026ndash;19 years in Bangladesh using nationally representative BAHWS 2019\u0026ndash;20 data. We used various machine learning classifiers, such as Logistic Regression, SVM, KNN, Decision Tree, Random Forest, Neural Network, and XGBoost. Ensemble-based approaches outperformed conventional classifiers, with XGBoost demonstrating the strongest and most stable predictive performance, achieving an accuracy of 76%, precision of 81.5%, recall of 86.5%, and an AUC-ROC of 72.9% after addressing class imbalance with SMOTE.\u003c/p\u003e \u003cp\u003eThe findings indicate that modern FP use among married adolescents is primarily influenced by spousal co-residency, childbirth experience, decision-making authority for family planning, administrative division, age group, wealth status, and media exposure. Adolescents whose husbands were residing outside Bangladesh and those who had not yet given birth were substantially less likely to use modern FP methods. This study also revealed that adolescent girls from the poorest households and those with no exposure to newspapers or magazines were less likely to use modern FP methods. In contrast, adolescents who made joint family planning decisions with their husbands, belonged to wealthier households, and were exposed to newspapers or magazines were more likely to use modern family planning. In contrast, adolescents who engaged in joint decision-making with their husbands, were in the younger age group (15\u0026ndash;17 years), and belonged to wealthier households, were significantly more likely to use modern FP. Geographic disparities were also evident, with higher FP use observed in Mymensingh and Rajshahi and lower use in Chittagong, Khulna, and Dhaka.\u003c/p\u003e \u003cp\u003eThe consistency among machine learning, SHAP analysis, and multivariable logistic regression strengthens the validity of these findings and demonstrates that ensemble machine learning models, particularly XGBoost, are effective at identifying specific risk factors among adolescents using modern family planning and provide valuable evidence for targeted interventions. Policymakers and program planners should consider these findings when designing adolescent-focused family planning strategies, with particular attention to adolescents affected by spousal separation, limited decision-making power, and regional and socioeconomic inequalities. Strengthening targeted approaches is essential to expanding equitable access to modern family planning services and achieving Sustainable Development Goal (SDG) target 3.7 in Bangladesh.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study used secondary, publicly available data. Ethical approval for the original survey was obtained from the relevant authorities. Therefore, additional ethical approval and informed consent were not required for this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data utilized in the current research can be accessed without registration at the UNC Dataverse (https://dataverse.unc.edu/).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have declared that no competing interests exist.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors’ contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization: Shawkatul Islam, Muhammad Khairul Alam\u003c/p\u003e\n\u003cp\u003eData curation: Shawkatul Islam\u003c/p\u003e\n\u003cp\u003eFormal analysis: Shawkatul Islam\u003c/p\u003e\n\u003cp\u003eMethodology: Shawkatul Islam, Md Eyah Eya,\u0026nbsp;Muhammad Khairul Alam\u003c/p\u003e\n\u003cp\u003eSoftware: Md. Rayhan Kabir, Md Eyah Eya\u003c/p\u003e\n\u003cp\u003eSupervision: Muhammad Khairul Alam\u003c/p\u003e\n\u003cp\u003eValidation: Muhammad Khairul Alam\u003c/p\u003e\n\u003cp\u003eVisualization: Shawkatul Islam\u003c/p\u003e\n\u003cp\u003eWriting – original draft: Shawkatul Islam, Md Eyah Eya\u003c/p\u003e\n\u003cp\u003eWriting – review \u0026amp; editing: Muhammad Khairul Alam\u003c/p\u003e\n\u003cp\u003eAll authors approved the final version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors are grateful to the National Institute of Population Research and Training (NIPORT), the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b), and Data for Impact (D4I) for their assistance in carrying out the Bangladesh Adolescent Health and Wellbeing Survey (BAHWS) 2019-20. We also like to thank the Bangladeshi government, the United States Agency for International Development (USAID)/Bangladesh, and the UK Foreign, Commonwealth, and Development Office (FCDO) for their assistance with the survey and for making the dataset publicly available.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors’ information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eFestin MPR. Overview of modern contraception. Best Pract Res Clin Obstet Gynaecol. 2020;66: 4\u0026ndash;14. doi:10.1016/j.bpobgyn.2020.03.004\u003c/li\u003e\n\u003cli\u003eTsui AO, McDonald-Mosley R, Burke AE. Family planning and the burden of unintended pregnancies. Epidemiol Rev. 2010;32: 152\u0026ndash;174. doi:10.1093/EPIREV/MXQ012\u003c/li\u003e\n\u003cli\u003eYehuala TZ. Exploring machine learning algorithms to predict not using modern family planning methods among reproductive age women in East Africa. BMC Health Services Research 2024 24:1. 2024;24: 1595-. doi:10.1186/S12913-024-11932-X\u003c/li\u003e\n\u003cli\u003eJR B, BJ S, AO L. Improving Birth Outcomes: Meeting the Challenge in the Developing World. Improving Birth Outcomes. 2003 [cited 31 Jan 2026]. doi:10.17226/10841\u003c/li\u003e\n\u003cli\u003eFamily planning/contraception methods. [cited 31 Jan 2026]. Available: https://www.who.int/news-room/fact-sheets/detail/family-planning-contraception\u003c/li\u003e\n\u003cli\u003eJimmy E, OsonwaKalu O, Nelson O, Dominic O. Prevalence of Contraceptive use among women of reproductive age in Calabar Metropolis, Southern Nigeria. 2013. \u003c/li\u003e\n\u003cli\u003eWorld Family Planning 2022 Meeting the changing needs for family planning: Contraceptive use by age and method. \u003c/li\u003e\n\u003cli\u003eNIPORT. National Institute of Population Research and Training (NIPORT), Mitra and Associates, \u0026amp; ICF International. (2023). Bangladesh Demographic and Health Survey 2022\u0026ndash;23: Key Indicators Report. Dhaka, Bangladesh, and Rockville, Maryland, USA: NIPORT, Mitra and Associates, and ICF International. 2023. \u003c/li\u003e\n\u003cli\u003eIslam AZ, Rahman M, Mostofa MG. Association between contraceptive use and socio-demographic factors of young fecund women in Bangladesh. Sex Reprod Healthc. 2017;13: 1\u0026ndash;7. doi:10.1016/J.SRHC.2017.05.001\u003c/li\u003e\n\u003cli\u003ePopulation Reference Bureau. World population data sheet. Population Reference Bureau Washington. 2015. . 2015. \u003c/li\u003e\n\u003cli\u003eKamal N. Contraceptive use among married adolescent girls in Bangladesh. J Biosoc Sci. 2013;45: 71\u0026ndash;86. \u003c/li\u003e\n\u003cli\u003eShahabuddin ASM, N\u0026ouml;stlinger C, Delvaux T, Sarker M, Bardaj\u0026iacute; A, Brouwere V De, et al. What Influences Adolescent Girls\u0026rsquo; Decision-Making Regarding Contraceptive Methods Use and Childbearing? A Qualitative Exploratory Study in Rangpur District, Bangladesh. PLoS One. 2016;11: e0157664. doi:10.1371/journal.pone.0157664\u003c/li\u003e\n\u003cli\u003eKamrul Islam M, Rabiul Haque M, Hema PS. Regional variations of contraceptive use in Bangladesh: A disaggregate analysis by place of residence. PLoS One. 2020;15. doi:10.1371/journal.pone.0230143\u003c/li\u003e\n\u003cli\u003eIslam AZ, others. Association between contraceptive use and socio-demographic factors of young fecund women in Bangladesh. Sex Reprod Healthc. 2017;13: 88\u0026ndash;94. \u003c/li\u003e\n\u003cli\u003eSDG Target 3.7 Sexual and reproductive health. [cited 31 Jan 2026]. Available: https://www.who.int/data/gho/data/themes/topics/sdg-target-3_7-sexual-and-reproductive-health\u003c/li\u003e\n\u003cli\u003eHaq I, Sakib S, Talukder A. Sociodemographic Factors on Contraceptive Use among Ever-Married Women of Reproductive Age: Evidence from Three Demographic and Health Surveys in Bangladesh. Medical Sciences. 2017;5: 31. doi:10.3390/medsci5040031\u003c/li\u003e\n\u003cli\u003eHossain M, Khan M, Ababneh F, Shaw J. Identifying factors influencing contraceptive use in Bangladesh: Evidence from BDHS 2014 data. BMC Public Health. 2018;18. doi:10.1186/s12889-018-5098-1\u003c/li\u003e\n\u003cli\u003eKundu S, Kundu S, Rahman MA, Kabir H, Al Banna MH, Basu S, et al. Prevalence and determinants of contraceptive method use among Bangladeshi women of reproductive age: a multilevel multinomial analysis. BMC Public Health. 2022;22. doi:10.1186/s12889-022-14857-4\u003c/li\u003e\n\u003cli\u003eHossain MI, others. Performance evaluation of machine learning algorithm for classification of unintended pregnancy among married women in Bangladesh. J Healthc Eng. 2022;2022: 1460908. \u003c/li\u003e\n\u003cli\u003eSidey-Gibbons JAM, Sidey-Gibbons CJ. Machine learning in medicine: a practical introduction. BMC Med Res Methodol. 2019;19: 64. doi:10.1186/s12874-019-0681-4\u003c/li\u003e\n\u003cli\u003eKebede SD, Sebastian Y, Yeneneh A, Chanie AF, Melaku MS, Walle AD. Prediction of contraceptive discontinuation among reproductive-age women in Ethiopia using Ethiopian Demographic and Health Survey 2016 Dataset: A Machine Learning Approach. BMC Med Inform Decis Mak. 2023;23. doi:10.1186/s12911-023-02102-w\u003c/li\u003e\n\u003cli\u003eWalle AD, Kebede SD, Adem JB, Mamo DN. Machine-learning algorithm to predict home delivery after antenatal care visit among reproductive age women in East Africa. Front Glob Womens Health. 2025;6. doi:10.3389/fgwh.2025.1461475\u003c/li\u003e\n\u003cli\u003eYehuala TZ. Exploring machine learning algorithms to predict not using modern family planning methods among reproductive age women in East Africa. BMC Health Serv Res. 2024;24: 1595. \u003c/li\u003e\n\u003cli\u003eMelaku MS, Yohannes L, Sharew B, Derseh MH, Taye EA. Application of machine learning algorithms to model predictors of informed contraceptive choice among reproductive age women in six high fertility rate sub Sahara Africa countries. BMC Public Health. 2025;25. doi:10.1186/s12889-025-23242-w\u003c/li\u003e\n\u003cli\u003eHossain MI, Habib MJ, Saleheen AAS, Kamruzzaman M, Rahman A, Roy S, et al. Performance Evaluation of Machine Learning Algorithm for Classification of Unintended Pregnancy among Married Women in Bangladesh. J Healthc Eng. 2022;2022. doi:10.1155/2022/1460908\u003c/li\u003e\n\u003cli\u003eChen T, Guestrin C. XGBoost: A scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2016. pp. 785\u0026ndash;794. doi:10.1145/2939672.2939785\u003c/li\u003e\n\u003cli\u003eBreiman L. Random Forests. 2001. \u003c/li\u003e\n\u003cli\u003eGupta S, Saluja K, Goyal A, Vajpayee A, Tiwari V. Comparing the performance of machine learning algorithms using estimated accuracy. Measurement: Sensors. 2022;24. doi:10.1016/j.measen.2022.100432\u003c/li\u003e\n\u003cli\u003eSani J, Halane S, Ahmed AM, Ahmed MM. Application of machine learning algorithms and SHAP explanations to predict fertility preference among reproductive women in Somalia. Sci Rep. 2025;15. doi:10.1038/s41598-025-04704-y\u003c/li\u003e\n\u003cli\u003e(NIPORT) NI of PR and T, International Centre for Diarrhoeal Disease Research b) B (icddr, Impact D for. Bangladesh Adolescent Health and Wellbeing Survey 2019-20. NIPORT, icddr, b, and Data for Impact Dhaka, Bangladesh, and Chapel Hill, NC \u0026hellip;; 2021. \u003c/li\u003e\n\u003cli\u003eWorld Health Organization. Reproductive Health and Research, World Health Organization. Medical eligibility criteria for contraceptive use, 5th ed. 2015: Geneva. 2015; 276. \u003c/li\u003e\n\u003cli\u003eKamrul Islam M, Rabiul Haque M, Hema PS. Regional variations of contraceptive use in Bangladesh: A disaggregate analysis by place of residence. PLoS One. 2020;15: e0230143. doi:10.1371/JOURNAL.PONE.0230143\u003c/li\u003e\n\u003cli\u003eAZ I, MN M, ML K, MM R, MR I, MG M, et al. Prevalence and Determinants of Contraceptive use among Employed and Unemployed Women in Bangladesh. Int J MCH AIDS. 2016;5. doi:10.21106/IJMA.83\u003c/li\u003e\n\u003cli\u003eRana MS, Khanam SJ, Alam MB, Hassen MT, Kabir MI, Khan MN. Exploration of modern contraceptive methods using patterns among later reproductive-aged women in Bangladesh. PLoS One. 2024;19: e0291100. doi:10.1371/JOURNAL.PONE.0291100\u003c/li\u003e\n\u003cli\u003eYang S, Berdine G. The receiver operating characteristic (ROC) curve. 2017;5: 34\u0026ndash;36. doi:10.12746/SWRCCC.V5I19.391\u003c/li\u003e\n\u003cli\u003eKursa MB, Rudnicki WR. Feature Selection with the Boruta Package. J Stat Softw. 2010;36: 1\u0026ndash;13. doi:10.18637/JSS.V036.I11\u003c/li\u003e\n\u003cli\u003eMahmud M, Islam MM. Adolescent contraceptive use and its determinants in Bangladesh: Evidence from Bangladesh Fertility Survey 1989. Contraception. 1995;52: 181\u0026ndash;186. doi:10.1016/0010-7824(95)00149-5\u003c/li\u003e\n\u003cli\u003eHuda F, Chowdhuri S, Sarker B, Islam N, Ahmed A. Prevalence of unintended pregnancy and needs for family planning among married adolescent girls living in urban slums of Dhaka, Bangladesh. 2014. doi:10.31899/rh4.1050\u003c/li\u003e\n\u003cli\u003eFeeser K, Chakraborty NM, Calhoun L, Speizer IS. Measures of family planning service quality associated with contraceptive discontinuation: an analysis of Measurement, Learning \u0026amp;amp; Evaluation (MLE) project data from urban Kenya. Gates Open Res. 2020;3: 1453. doi:10.12688/gatesopenres.12974.2\u003c/li\u003e\n\u003cli\u003eVan Den Goorbergh R, Van Smeden M, Timmerman D, Ben Van Calster. The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression. Journal of the American Medical Informatics Association. 2022;29: 1525\u0026ndash;1534. doi:10.1093/JAMIA/OCAC093\u003c/li\u003e\n\u003cli\u003eAdem JB, Kebede SD, Walle AD, Mamo DN. Predicting determinants of modern contraceptive use among reproductive-age women in Ethiopia using machine learning algorithm: Evidence from the Performance Monitoring and Accountability (PMA) Survey 2019 dataset. F1000Research 2025 14:99. 2025;14: 99. doi:10.12688/f1000research.156316.1\u003c/li\u003e\n\u003cli\u003eSani J, Oluwagbemiga A, Ahmed MM. Machine learning-based prediction of optimal antenatal care utilization among reproductive women in Nigeria. Machine Learning with Applications. 2025;21: 100698. doi:10.1016/J.MLWA.2025.100698\u003c/li\u003e\n\u003cli\u003eTadese ZB, Nimani TD, Mare KU, Gubena F, Wali IG, Sani J. Exploring machine learning algorithms for predicting fertility preferences among reproductive age women in Nigeria. Front Digit Health. 2025;6: 1495382. doi:10.3389/FDGTH.2024.1495382\u003c/li\u003e\n\u003cli\u003eLundberg S, Lee S-I. A Unified Approach to Interpreting Model Predictions. 2017. Available: http://arxiv.org/abs/1705.07874\u003c/li\u003e\n\u003cli\u003eRois R, Ray M, Rahman A, Roy SK. Prevalence and predicting factors of perceived stress among Bangladeshi university students using machine learning algorithms. Journal of Health, Population and Nutrition 2021 40:1. 2021;40: 50-. doi:10.1186/S41043-021-00276-5\u003c/li\u003e\n\u003cli\u003eKhan R, MacQuarrie KLD, Sultana M, Nahar Q. Intermittent Needs for Family Planning among Women with an Internal Migrant Husband in Bangladesh: A Qualitative Study. Sex Reprod Health Matters. 2022;29: 2097044. doi:10.1080/26410397.2022.2097044\u003c/li\u003e\n\u003cli\u003eSamanta R, Munda J. Husband\u0026rsquo;s migration status and contraceptive behaviors of women: evidence from Middle-Ganga Plain of India. BMC Womens Health. 2023;23: 180. doi:10.1186/S12905-023-02325-Z\u003c/li\u003e\n\u003cli\u003eTesfa GA, Demeke AD, Seboka BT, Tebeje TM, Kasaye MD, Gebremeskele BT, et al. Employing machine learning models to predict pregnancy termination among adolescent and young women aged 15\u0026ndash;24 years in East Africa. Scientific Reports 2024 14:1. 2024;14: 30047-. doi:10.1038/s41598-024-81197-1\u003c/li\u003e\n\u003cli\u003ePradhan MR, Mondal S. Examining the influence of Mother-in-law on family planning use in South Asia: insights from Bangladesh, India, Nepal, and Pakistan. BMC Women\u0026rsquo;s Health 2023 23:1. 2023;23: 418-. doi:10.1186/s12905-023-02587-7\u003c/li\u003e\n\u003cli\u003eKhan MN, Khan MMA, Billah MA, Khanam SJ, Haider MM, Sarker BK, et al. Effects of maternal healthcare service utilization on modern postpartum family planning access in Bangladesh: insights from a National representative survey. PLoS One. 2025;20: e0318363. doi:10.1371/JOURNAL.PONE.0318363\u003c/li\u003e\n\u003cli\u003eIslam MM, Islam MK, Hasan MS, Hossain MB. Adolescent motherhood in Bangladesh: Trends and determinants. PLoS One. 2017;12: e0188294. doi:10.1371/JOURNAL.PONE.0188294\u003c/li\u003e\n\u003cli\u003eRahman MM, Mostofa MG, Hoque MA. Women\u0026rsquo;s household decision-making autonomy and contraceptive behavior among Bangladeshi women. Sexual \u0026amp; Reproductive Healthcare. 2014;5: 9\u0026ndash;15. doi:10.1016/J.SRHC.2013.12.003\u003c/li\u003e\n\u003cli\u003eHameed W, Azmat SK, Ali M, Sheikh MI, Abbas G, Temmerman M, et al. Women\u0026rsquo;s Empowerment and Contraceptive Use: The Role of Independent versus Couples\u0026rsquo; Decision-Making, from a Lower Middle Income Country Perspective. PLoS One. 2014;9: e104633. doi:10.1371/JOURNAL.PONE.0104633\u003c/li\u003e\n\u003cli\u003eIslam AZ. Factors affecting modern contraceptive use among fecund young women in Bangladesh: does couples\u0026rsquo; joint participation in household decision making matter? Reprod Health. 2018;15: 112. doi:10.1186/S12978-018-0558-8\u003c/li\u003e\n\u003cli\u003eAlam N, Mollah MMH, Naomi SS. Prevalence and determinants of adolescent childbearing: comparative analysis of 2017\u0026ndash;18 and 2014 Bangladesh Demographic Health Survey. Front Public Health. 2023;11: 1088465. doi:10.3389/FPUBH.2023.1088465/FULL\u003c/li\u003e\n\u003cli\u003eAkinyemi AI, Ikuteyijo OO, Mobolaji JW, Erinfolami T, Adebayo SO. Socioeconomic inequalities and family planning utilization among female adolescents in urban slums in Nigeria. Front Glob Womens Health. 2022;3: 838977. doi:10.3389/FGWH.2022.838977\u003c/li\u003e\n\u003cli\u003eYesuf KA, Liyew AD, Bezabih AK. Impact of exposure to mass media on utilization modern contraceptive among adolescent married women in Ethiopia: evidence from Ethiopia demographic health survey 2016. International Journal of Scientific Reports. 2021;7: 434. doi:10.18203/issn.2454-2156.IntJSciRep20213257\u003c/li\u003e\n\u003cli\u003eMoon MP. Contraceptive behaviors and media influence among women in Bangladesh: exploring the effects of age and education. Front Glob Womens Health. 2025;6: 1492105. doi:10.3389/fgwh.2025.1492105\u003c/li\u003e\n\u003cli\u003ePlus S. Sources of Family Planning Bangladesh. 2016. \u003c/li\u003e\n\u003cli\u003eKhan MHR, Siddik AB, Islam T. Disparities in contraceptive preferences among Bangladeshi women: a multilevel logistic regression study. BMC Public Health. 2025;25: 3444. doi:10.1186/s12889-025-24424-2\u003c/li\u003e\n\u003cli\u003eKamrul Islam M, Rabiul Haque M, Hema PS. Regional variations of contraceptive use in Bangladesh: A disaggregate analysis by place of residence. PLoS One. 2020;15: e0230143. doi:10.1371/journal.pone.0230143\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-public-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pubh","sideBox":"Learn more about [BMC Public Health](http://bmcpublichealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pubh/default.aspx","title":"BMC Public Health","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Modern family planning, married adolescent, machine learning, SHAP analysis, Bangladesh","lastPublishedDoi":"10.21203/rs.3.rs-8806294/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8806294/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground:\u003c/h2\u003e \u003cp\u003eModern family planning (FP) is essential for improving the reproductive health of adolescent women, especially in Bangladesh, where early marriage is still prevalent. Despite overall progress in family planning, modern FP use remains uneven and relatively low among married adolescent women due to social, relational, and contextual barriers. Conventional statistical approaches might not fully capture the intricate, nonlinear determinants of FP use. Therefore, this study applied and compared multiple machine learning (ML) models to predict modern FP use and identify its key determinants among married adolescent women in Bangladesh.\u003c/p\u003e\u003ch2\u003eMethods and materials:\u003c/h2\u003e \u003cp\u003eData were obtained from the nationally representative Bangladesh Adolescent Health and Wellbeing Survey (BAHWS) 2019\u0026ndash;20. The analysis included a weighted sample of 3,223 ever-married adolescent females aged 15 to 19 years. A variety of machine learning classification models were employed, including Logistic Regression (LR), Random Forest (RF), Decision Tree (DT), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Neural Network (NN), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Adaptive Boosting (AdaBoost), and Categorical Na\u0026iuml;ve Bayes (CNB). Feature selection was performed with the Boruta algorithm, and model interpretability was examined through SHapley Additive exPlanations (SHAP). Model performance was assessed using accuracy, precision, recall, F1-score, Matthews Correlation Coefficient, Cohen's Kappa, and area under the receiver operating characteristic curve (AUROC).\u003c/p\u003e\u003ch2\u003eResult:\u003c/h2\u003e \u003cp\u003eOverall, 72.7% of married adolescent women reported using a modern FP method. Ensemble-based models outperformed conventional classifiers, especially after class balancing. XGB had the best overall predictive performance after applying SMOTE, with 76.0% accuracy, 81.5% precision, 83.9% F1 Score, 37.2% MCC, 36.9% Cohen's Kappa, 86.5% recall, and an AUROC of 72.9%, followed by RF and NN models. The most important determinants of modern FP use included spousal co-residency, having given birth, joint decision-making about FP use, administrative division, younger age (15\u0026ndash;17), household wealth, and media exposure.\u003c/p\u003e\u003ch2\u003eConclusion:\u003c/h2\u003e \u003cp\u003eThis study reveals that machine learning models, especially XGBoost, can be useful in predicting modern family planning use and identifying key determinants among married adolescent women in Bangladesh. The findings provide a data-driven foundation for policymakers and program planners to create adolescent-focused family planning programs. The results highlight the significance of regional disparities, socioeconomic inequities, and relationship factors in influencing adolescent contraceptive behavior. Integrating ML-based evidence into family planning programs may enhance targeted interventions and help to achieve Sustainable Development Goal 3.7.\u003c/p\u003e","manuscriptTitle":"Determinants of modern family planning use among married adolescents in Bangladesh: Evidence from a machine learning approach","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-12 17:06:38","doi":"10.21203/rs.3.rs-8806294/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-30T14:49:56+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"215739115569758134631456661829717725610","date":"2026-03-15T12:06:04+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-14T19:39:16+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-13T18:17:27+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"312582823693888759435306059794499987856","date":"2026-03-13T17:56:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"104310510925182140421752137694388777018","date":"2026-03-11T15:52:30+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"48476881638282003369508038997208813660","date":"2026-03-08T19:35:30+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"232635740794753446728354842865202588412","date":"2026-03-06T17:31:35+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-06T15:44:39+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-04T11:32:39+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-02-11T18:22:52+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-11T07:00:03+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Public Health","date":"2026-02-11T06:45:02+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-public-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pubh","sideBox":"Learn more about [BMC Public Health](http://bmcpublichealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pubh/default.aspx","title":"BMC Public Health","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"9eaa47e6-9bb6-4ff9-b50a-262dde375bf4","owner":[],"postedDate":"March 12th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-04T11:53:25+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-12 17:06:38","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8806294","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8806294","identity":"rs-8806294","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}