Ensemble Machine Learning with SHAP Interpretability for Predicting Unmet Contraceptive Needs in Ethiopia

Ensemble Machine Learning with SHAP Interpretability for Predicting Unmet Contraceptive Needs in Ethiopia | 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 Ensemble Machine Learning with SHAP Interpretability for Predicting Unmet Contraceptive Needs in Ethiopia Melaku Alelign Mengstie, Alexander Takele Mengesha This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7454492/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Unmet contraceptive needs remain a critical challenge in global reproductive health, especially in developing countries like Ethiopia, where access to family planning is limited. Women who wish to delay or avoid pregnancy but do not use contraception face increased risks of unintended pregnancies, unsafe abortions, and adverse maternal outcomes. This study aims to develop and evaluate an ensemble machine learning model, enhanced with Explainable AI techniques, to accurately identify women at risk of unmet contraceptive needs, thereby supporting informed and transparent decision-making. Data from the 2011 and 2016 Ethiopian Demographic and Health Surveys were used. Ensemble models, including Random Forest, Categorical Boosting, Extreme Gradient Boosting, and Light Gradient Boosting Machine, were trained on 21 key features selected through Recursive Feature Elimination. A hybrid SMOTE-Tomek sampling technique addressed class imbalance. Stratified train-validation-test splits ensured robust performance evaluation. Extreme Gradient Boosting emerged as the best-performing model, achieving 96.56% accuracy, 97.59% precision, 95.99% recall, and a 96.53% F1-score in cross-validation. On the test set, it maintained strong results with 95.55% accuracy and a 90.90% F1-score, outperforming Logistic Regression and Support Vector Machine. Receiver Operating Characteristic curve analysis confirmed its excellent classification (AUC = 0.99). SHapley Additive exPlanations analysis highlighted key predictors driving Extreme Gradient Boosting’s predictions, including contraceptive information exposure, prior family planning use, pregnancy intention, decision-making autonomy, and fertility preferences. The findings provide interpretable, data-driven insights for targeted reproductive health interventions. Integrating such predictive models into real-time health systems may enhance family planning strategies and help achieve Ethiopia’s 2030 health goals. Ensemble learning Explainable AI Family planning Machine learning Reproductive health SHAP values Unmet contraceptive needs XGBoost Figures Figure 1 Figure 2 Figure 3 Key Messages What is already known about this topic Unmet contraceptive needs persist in Ethiopia, and previous predictive models lacked accuracy and interpretability. What this study adds This study applies an ensemble machine learning approach enhanced with SHAP to predict unmet contraceptive needs, identifying key factors like contraceptive awareness and fertility preferences. How might this study affect research, practice, or policy The model’s transparency and high accuracy can guide targeted reproductive health interventions and support data-driven policy aligned with SDG 3.7. 1. INTRODUCTION Family planning plays a crucial role in public health by enabling individuals to control the timing and number of children. This improves maternal and child health, reduces unintended pregnancies, and supports women's education and careers. However, unmet contraceptive needs remain a pressing issue, particularly in developing regions[1]. Unmet contraceptive needs when individuals who wish to delay or prevent pregnancy are not using any method of contraception pose a persistent public health challenge[2]. Over 218 million women globally face an unmet need for family planning, contributing to a high rate of unintended pregnancies due to limited use of modern contraceptive methods[3]. In sub-Saharan Africa, unintended pregnancies remain widespread, influenced by factors such as education, contraceptive awareness, marital status, and regional disparities[4]. Unintended pregnancies lead to serious health risks, including preterm labour, unsafe abortions, and maternal depression[5][6]. In Ethiopia, A significant proportion of women express a desire to space or limit births but still lack access to appropriate contraceptive methods. National surveys have reported high rates of unintended pregnancies and persistent regional and socioeconomic disparities in contraceptive use[7]. Despite progress in long-acting and permanent contraceptive use, rising from 17.9% in 2014 to 35.5% in 2019, the contraceptive prevalence rate is still far below the 2020 target of 55%[8]. Ethiopia’s Federal Ministry of Health aims to reduce the unmet need for family planning from 22% to 17% [9] and increase the contraceptive prevalence rate to 54% by 2030[10][11]. Data-driven approaches are vital for effectively identifying women at risk of unmet contraceptive needs. Despite efforts, many face barriers to access, partly due to a lack of accurate predictive tools. Ensemble machine learning offers precise risk prediction, helping policymakers and healthcare providers target interventions better. Improving contraceptive prevalence and reducing unmet need align with 2030 Sustainable Development Goal 3.7 to ensure universal access to sexual and reproductive health services[12]. Machine Learning Algorithms Machine Learning (ML) algorithms learn patterns from socio-demographic, behavioral, and health data to capture complex, non-linear relationships often missed by traditional statistics[13]. Ensemble methods, Random Forest, CatBoost, LightGBM, and XGBoost combine multiple weak learners to enhance accuracy and robustness[14]; CatBoost excels with categorical data and overfitting prevention[15]; LightGBM accelerates training with leaf-wise splits[16] ; XGBoost uses gradient optimization and regularization[17]. and Random Forest improves generalization by aggregating many decision trees[18]. In family planning, ML predicts unmet contraceptive needs, evaluates programs, identifies access barriers, and forecasts demand. When integrated with mobile health platforms, ML empowers personalized reproductive health decisions[19][20]. Explainable AI (XAI) techniques like SHAP enhance transparency by clarifying feature influence, enabling data-driven, equitable policy interventions. Deploying ML in Ethiopia could optimize family planning resources, improve service delivery, and help reach a 54% contraceptive prevalence rate by 2030, reducing unintended pregnancies and improving maternal health. This study develops an ensemble machine-learning model to predict unmet contraceptive needs among Ethiopian reproductive-age women. By integrating multiple algorithms, it aims to enhance prediction accuracy and identify key factors that drive unmet needs. The findings will inform policies and interventions aimed at enhancing family planning services in Ethiopia. The following questions guide the research: What are the relevant features for constructing the predictive model for unmet contraceptive needs?Which ensemble machine learning algorithm is most suitable for predicting unmet contraceptive needs among Ethiopian reproductive-age women? How can Explainable AI techniques be applied to enhance the interpretability of the predictive model, and in identifying the most influential factors influencing unmet contraceptive needs? II. Related Works D. Mamo et.al [21]predicted unintended pregnancy using six machine learning algorithms on 7,193 secondary data points from the 2016 EDHS survey. Extra Trees emerged as the top model, achieving 92.8% AUC with default hyperparameter tuning, identifying predictors such as region, ideal number of children, and education. However, ensemble methods were not utilized, prompting our focus on advanced ensemble algorithms and XAI for unmet contraceptive needs in Ethiopia. S. D. Kebede et.al [22]used machine learning to predict the unmet contraceptive needs of 5,819 women from the 2019 PMA survey, identifying Random Forest as the best model (85% accuracy, 0.93 AUC). SHAP revealed key predictors like partner disapproval and regional factors, but ensemble techniques were omitted, highlighting the need for our advanced approach to integrating XAI. S. Kebede et al [23]explored contraceptive discontinuation using Random Forest (68% accuracy) on EDHS data, identifying age, education, and family size as key predictors. However, ensemble techniques and XAI were absent, necessitating our study’s innovative approach to improve predictions and transparency. Similarly, T. Yehuala [24] applied XGBoost to predict non-use of modern family planning among 92,564 East African women, identifying education, age, rural residence, marital status, contraceptive knowledge, and smoking as key predictors. Similarly, S. Kassie et al. [25] used Random Forest to predict community-based health insurance enrollment among 9,013 Ethiopian women, highlighting residence, wealth, household head age, husband’s education, media exposure, family size, and number of under-five children as influential factors. While both studies demonstrate the power of machine learning for health prediction, they lacked systematic multi-algorithm benchmarking and interpretable, policy-focused insights. Our study addresses these gaps by combining ensemble algorithms with SHAP to provide accurate and explainable predictions specific to unmet contraceptive needs. Previous studies [21]to[23]provided valuable insights but lacked the innovation of ensemble learning and Explainable AI (XAI). Our research addresses this gap by integrating advanced ensemble algorithms with XAI, enhancing predictive accuracy, interpretability, and policy relevance. Similarly, T. Z. Yehuala [24] and S. Y. Kassie et al. [25] applied machine learning to reproductive health outcomes, but did not focus on unmet contraceptive needs or leverage ensemble methods with XAI. Our study fills this gap, delivering more reliable and interpretable predictions of unmet contraceptive needs in Ethiopia, supporting targeted family planning interventions. Provided valuable insights but lacked the innovation of ensemble learning and Explainable AI. Our research addresses this gap by integrating advanced ensemble algorithms with XAI, enhancing predictive accuracy, interpretability, and policy relevance. Similarly,[24] and [25] applied machine learning to reproductive health outcomes, but did not focus on unmet contraceptive needs. Our study fills this gap, delivering more reliable and interpretable predictions of unmet contraceptive needs in Ethiopia, supporting targeted family planning interventions. III. Materials and Methods Figure 1 illustrates the workflow from raw data to actionable insights. It covers data cleaning and transformation, balancing with SMOTETomek, selecting key features, training multiple models, and using SHAP to interpret the results. A. Data Collection The researchers analyzed data from the Ethiopian Demographic and Health Surveys (2011–2016), which included socio-demographic, behavioral, and health-related information. This comprehensive dataset allowed them to explore patterns and factors influencing contraceptive access and use across different populations and regions in Ethiopia. B. Data preprocessing After collecting the secondary dataset from EDHS, the next step is data preprocessing, which includes data cleaning, transformation, feature selection, class balancing, data splitting, and other necessary steps. 1. Data Cleaning Missing numerical values such as family size, husband’s age, and child’s birth size were filled using rounded means (5, 37, 41). For important categorical variables like pregnancy intention, contraceptive decision-making, and HIV testing method, imputation maintained the original data distribution, ensuring reliable analysis of reproductive and socio-demographic factors. 2. Data Transformation With guidance from domain experts, continuous variables such as family size, birth interval, antenatal visits, and BMI were grouped into meaningful categories to improve model interpretability. Key categorical features, including place of delivery, water source, and housing materials, were recoded for consistency. These expert-informed transformations aligned the data with real-world health contexts, supporting more effective modeling. C. Data Splitting The dataset was split into 20% test data, with the remaining 80% further divided into training and validation sets (80:20). This ensured consistent class distribution across all subsets for reliable model evaluation. 3. Feature Selection We used Recursive Feature Elimination to determine the optimal threshold for feature selection, which resulted in identifying 21 key features. By iteratively eliminating less important variables, RFE sets this threshold and retains only the most relevant predictors for subsequent model development, thereby enhancing the model’s efficiency and effectiveness. 4. Addressing Class Imbalance To handle class imbalance in the training set, SMOTETomek was used, combining oversampling of the minority class with under-sampling through Tomek links. This balanced the ‘unmet need’ classes to 8,410 each, reducing bias toward the majority. Applied only to the training data, it preserved the validation and test sets, ensuring reliable model evaluation (see Supplemental Figure 1). D. Model Development Multiple models were developed to predict unmet contraceptive needs. Ensemble methods Random Forest, CatBoost, XGBoost, and LightGBM were used alongside Logistic Regression and SVM as benchmarks, allowing comparison between more complex and simpler approaches. E. Model Validation and Evaluation Hyperparameters were tuned, and stratified five-fold cross-validation ensured model stability. The final evaluation on the test set used accuracy, precision, recall, F1 score, and AUC-ROC. SHAP analysis was applied to interpret the models and identify key predictive features. IV. Results and Discussion A. Experimental Result Analysis of Relevant Features Using Recursive Feature Elimination with a Random Forest classifier, 21 key features were selected from 88, prominent a balance between accuracy and simplicity. These features highlighted the most important demographic, reproductive, behavioral, and health-related factors. Leading predictors included contraception exposure, fertility desire, pregnancy intention, intention to use, and past family planning use. Other influential factors, such as region, wealth index, contraceptive decision-making, HIV testing, and partner’s education, further improved model performance and interpretability (see Supplemental Figure 2).Top of FormBottom of Form Predictive Model Development Trained on 21 key features selected via RFE, the models captured complex patterns across demographic, socioeconomic, healthcare, and reproductive domains. Performance metrics from stratified five-fold cross-validation are shown in Table 1, with final test results in Table 2. Table 1 Five-Fold Cross-Validation Performance Model Accuracy Precision Recall F1 Score Random Forest 96.35 97.90 94.72 96.28 CatBoost 96.52 97.44 95.95 96.69 XGBoost 96.56 97.59 95.99 96.53 LightGBM 96.42 96.53 95.62 96.52 Logistic Regression 86.37 86.46 86.38 86.36 Support Vector Machine 93.22 93.24 93.22 93.21 Table 2 Performance of Selected Ensemble Algorithms using Normal Test Set Model Accuracy Precision Recall F1 Score Random Forest 95.44 93.54 87.73 90.63 CatBoost 95.01 91.88 87.95 89.87 XGBoost 95.55 93.67 88.29 90.90 LightGBM 95.21 91.95 88.74 90.32 Logistic Regression 83.68 63.88 80.86 71.37 Support Vector Machine 92.58 84.78 85.92 85.35 Best Model Performance and Selection XGBoost proved to be the top model for predicting unmet contraceptive needs, achieving 95.55% accuracy, 93.67% precision, 88.29% recall, and a 90.90% F1 score on the test set. It consistently outperformed other models in cross-validation by balancing sensitivity and reducing false positives. As shown in Supplemental Figure 3, ROC analysis indicated that XGBoost, Random Forest, and LightGBM reached near-perfect AUCs of 0.99, demonstrating the superior predictive power of tree-based ensembles compared to traditional methods like Logistic Regression and SVM. These findings establish XGBoost as the most reliable and effective model for this task. Discussion 1. What are the relevant features for constructing the predictive model and the potential risk factors for unmet contraceptive needs? As presented in Supplemental Figure 2, the top 21 features selected through Recursive Feature Elimination represent the key risk factors associated with unmet contraceptive needs in Ethiopia. These features highlight the critical role of exposure to contraceptive information, previous use of family planning methods, pregnancy intention, intention to use contraception, and current pregnancy status. Fertility preferences, including the desire for more children and the ideal number of children, along with awareness of alternative family planning methods and decision-making autonomy, also emerge as important predictors. Together, these variables illustrate the complex relationships among reproductive intentions, access to information, personal agency, and healthcare engagement. This selection of features provides valuable guidance for developing targeted, data-driven reproductive health interventions aimed at reducing disparities and improving outcomes. Bottom of Form 2. Which ensemble machine-learning technique is most effective for building a predictive model for unmet contraceptive needs among Ethiopian women? XGBoost emerged as the most effective ensemble model, utilizing 21 key features identified through Recursive Feature Elimination. During five-fold cross-validation, it achieved a top precision of 97.59%, alongside strong recall and F1 scores, accurately identifying cases while minimizing false positives. On the test set, XGBoost outperformed all other models with 95.55% accuracy, 93.67% precision, 88.29% recall, and a 90.90% F1 score, demonstrating robustness and generalizability. Its strength lies in capturing complex, nonlinear patterns, handling missing data, and preventing overfitting through regularization critical for modeling Ethiopia’s diverse socioeconomic and reproductive factors. Beyond impressive performance metrics, XGBoost provides actionable insights, making it a powerful tool for policymakers and health professionals addressing unmet contraceptive needs. Bottom of Form 3. How can Explainable AI techniques be applied to enhance the interpretability of the predictive model, and in identifying the most influential factors influencing unmet contraceptive needs? Explainable AI techniques, such as SHAP, improve the transparency and interpretability of predictive models for unmet contraceptive needs. Using SHAP, this study provided both global and local explanations of the XGBoost model’s predictions. Figure 2 shows the average absolute SHAP values, highlighting the top eight influential features: exposure to contraception information, prior family planning use, desire for more children, intention to use contraception, pregnancy intention, current pregnancy status, knowledge of other family planning methods, and primary decision-maker on contraception. These factors strongly shape the model’s outcomes. As presented, Supplemental Figure 4 presents the SHAP summary plot illustrating both the magnitude and direction of each feature’s effect on unmet contraceptive needs, showing which features increase risk (e.g., desire for more children, pregnancy intention) and which decrease risk (e.g., exposure to contraception information, prior family planning use). At the individual level, SHAP waterfall plots (Figure 3) break down how features push predictions toward higher or lower risk, revealing personalized influences that may differ from global trends. This granular insight fosters ethical, informed public health decisions by explaining not just what the model predicts, but why. Our study confirms key risk factors for unmet contraceptive needs in Ethiopia, with exposure to contraceptive information as a critical determinant[2]. Consistent with prior research, lack of media exposure increases unmet need odds by 1.32 times in high-fertility areas. These results highlight improving access to reproductive health information as a vital strategy to reduce unmet contraceptive needs. Research in East Africa shows that a woman’s intention to use contraception strongly predicts uptake, with intention within a year linked to higher adoption rates[26]. In Ethiopia, greater decision-making autonomy reduces unmet contraceptive need, while unintended or current pregnancies are consistently associated with higher unmet need[27]. Additionally, women desiring more children are significantly more likely to report unmet need, highlighting the influence of fertility preferences on contraceptive use[28]. Key findings reveal that unmet contraceptive needs in Ethiopia are mainly driven by limited family planning information, low decision-making autonomy, and unclear fertility intentions. Women lacking exposure to contraceptive messaging and those with a strong desire for more children face higher risks. SHAP analysis confirms these factors as the most influential, emphasizing the critical role of empowerment and informed reproductive choices. Limitations This study, while insightful, has limitations. It relies on cross-sectional EDHS data, restricting causal and temporal analysis. Self-reported data may introduce recall bias, impacting accuracy. SHAP improves interpretability but may miss complex feature interactions and does not imply causation. Lastly, the model’s applicability outside Ethiopia is uncertain due to differing sociocultural and healthcare contexts. Future work should use longitudinal data and validate the model in diverse populations for greater robustness and policy relevance. V. Conclusion and Recommendation Conclusion Unmet contraceptive needs remain a major challenge in Ethiopia, driven by limited access to family planning information, low women’s autonomy, and unclear fertility intentions. These factors hinder effective contraceptive use and impact reproductive health outcomes. Using EDHS data, our study applied XGBoost and Explainable AI to predict unmet needs with high accuracy. Key predictors identified include pregnancy intention, prior contraceptive use, desire for more children, exposure to family planning information, and decision-making autonomy. SHAP analysis clarified the influence of these factors, enhancing model transparency. These findings underscore the urgent need for policies that expand access to reproductive health information and empower women’s decision-making. Integrating AI-driven predictive tools into health systems can enable targeted interventions, improving resource allocation and program effectiveness. Tailored counseling aligned with women’s fertility goals and broader efforts to promote gender equity are essential to reducing unmet contraceptive needs. Our work provides actionable insights to guide policymakers and health professionals in advancing reproductive health equity and achieving national and global family planning goals. Contribution The research advances public health and AI integration by developing a robust predictive framework employing ensemble algorithms (XGBoost, CatBoost, LightGBM, Random Forest), balanced via SMOTETomek and optimized through Recursive Feature Elimination. SHAP explainability ensures transparent outputs, empowering policymakers to focus on critical factors like contraceptive knowledge and women’s autonomy. Leveraging publicly available data, the scalable model offers a practical tool adaptable to similar low-resource settings for data-driven reproductive health strategies. Recommendation Based on our findings, we recommend integrating XGBoost with SHAP into national health planning to identify high-risk groups and improve resource allocation. Expanding targeted health education through community outreach, empowering women’s reproductive autonomy, and enhancing provider counseling are essential. Future research should use longitudinal and real-time data and test this framework in other sub-Saharan contexts. Successful implementation requires strong digital infrastructure, institutional support, and capacity building for sustainable, data-driven public health decisions. Declarations Ethics Approval and Consent to Participate Not applicable. Clinical Trial Number Not applicable. Consent for Publication Not applicable. Competing Interests The authors declare that they have no competing interests. Funding No funding was received for this research. Author Contribution Authors' Contributions•Melaku Alelign Mengstie: Conceptualization, data analysis, and manuscript writing.•Alexander Takele Mengesha: data collection, methodology, and Interpretation.All authors read and approved the final manuscript. Acknowledgement We sincerely thank the Ethiopian Demographic and Health Surveys for providing the essential data for this study. We also gratefully acknowledge the public health and reproductive health experts from the University of Gondar, whose guidance was vital in data processing, feature selection, and interpreting results, ensuring the study’s relevance to real-world challenges and policy Data Availability The dataset used in this study, the Ethiopian Demographic and Health Survey (EDHS), is publicly available. All data and materials related to the study can be accessed upon reasonable request from the corresponding author.The dataset is available at the EDHS website: [https://dhsprogram.com/data/available-datasets.cfm](https:/dhsprogram.com/data/available-datasets.cfm) References Smith M, Keckley P. Adding it up. Hosp Heal Networks. 2017;91(10):64–6. 10.1177/0162643418759341 . Asmamaw DB, Negash WD. Magnitude of unmet need for family planning and its predictors among reproductive age women in high fertility regions of Ethiopia: Evidence from Ethiopian Demographic and Health Survey. BMC Womens Health. 2022;22(1):1–10. 10.1186/s12905-022-01982-w . Nemser B, Addofoh N. Contextual factors associated with contraceptive utilization and unmet need among sexually active unmarried women in Kenya: A multilevel regression analysis, PLoS One , vol. 17, no. 6 June, pp. 1–21, 2022, 10.1371/journal.pone.0270516 Ayalew HG, et al. Prevalence and factors associated with unintended pregnancy among adolescent girls and young women in sub-Saharan Africa, a multilevel analysis. BMC Womens Health. 2022;22(1):1–9. 10.1186/s12905-022-02048-7 . Gelaw KA, Atalay YA, Gebeyehu NA. Unintended pregnancy and contraceptive use among women in low – and middle – income countries: systematic review and meta – analysis, pp. 1–13, 2023, 10.1186/s40834-023-00255-7 Garo MG, Abe SG, Girsha WD, Daka DW. Unmet need for family planning and associated factors among currently married women of reproductive age in Bishoftu town, Eastern Ethiopia. PLoS ONE. 2021;16(12):1–16. 10.1371/journal.pone.0260972 . Ethiopian Public Health Institute (EPHI) and ICF, Ethiopia Mini Demographic and Health Survey 2019: Final Report. 2021. [Online]. Available: https://dhsprogram.com/pubs/pdf/FR363/FR363.pdf Ethiopian Ministry of Health. Health Sector Transformation Plan II 2020/2021–2024/2025, Ethiop. Minist. Heal. , vol. 25, no. February, pp. 1–128, 2021. Vision P. Ethiopia ’ s FP2030 Commitment Approach to Accountability, no. 7200, 2025. UNFPA Ethiopia. | Ethiopia launches a costed family planning implementation plan. Accessed: Apr. 03, 2025. [Online]. Available: https://ethiopia.unfpa.org/en/news/ethiopia-launches-costed-family-planning-implementation-plan Shumet T, Geda NR, Hassan JA. Barriers to modern contraceptive utilization in Ethiopia. Contracept Reprod Med. 2024;9(1). 10.1186/s40834-024-00311-w . Nations U. World Contraceptive Use, https://www.un.org/development/desa/pd/data/world-contraceptive-use Taye SBM, et al. Predicting Unmet Contraceptive Needs Using Machine Learning Techniques. J Health Inf Developing Ctries. 2021;15(1):1–10. Series C. Extreme gradient boosting (XGBoost) method in making forecasting application and analysis of USD exchange rates against rupiah Extreme gradient boosting (XGBoost) method in making forecasting application and analysis of USD exchange rates against rupi, pp. 0–11, 2021, 10.1088/1742-6596/1722/1/012016 Tanha J, Abdi Y, Samadi N, Razzaghi N, Asadpour M. Boosting methods for multi-class imbalanced data classification: an experimental review. J Big Data. 2020;7(1). 10.1186/s40537-020-00349-y . Syam P, Chand S, Divya G. A Light Gradient Boosting Machine Regression Model for Prediction of Agriculture Insurance Cost over Linear Regression, pp. 200–8, 2022, 10.3233/APC220027 Ganaie MA, Hu M, Tanveer* M, Suganthan* PN. Ensemble deep learning: A review, 2021. Smirani LK, Yamani HA, Menzli LJ, Boulahia JA. Using Ensemble Learning Algorithms to Predict Student Failure and Enabling Customized Educational Paths, Sci. Program. , vol. 2022, 2022. 10.1155/2022/3805235 Obermeyer EJ, Z., Emanuel. Predicting the Future—Big Data, Machine Learning, and Clinical Medicine. N Engl J Med. 2016;375(13):1216–9. https://doi.org/10.1056/NEJMp1606181 . Classification HI. A Comprehensive Review on Machine Learning in Healthcare Industry: Classification, Restrictions, Opportunities and Challenges, 2023. Mamo DN, Gebremariam YH, Adem JB, Kebede SD, Walle AD. Machine learning to predict unintended pregnancy among reproductive-age women in Ethiopia: evidence from EDHS 2016. BMC Womens Health. 2024;24(1):1–11. 10.1186/s12905-024-02893-8 . Kebede SD, Mamo DN, Adem JB, Semagn BE, Walle AD. Machine learning modeling for identifying predictors of unmet need for family planning among married/in-union women in Ethiopia: Evidence from performance monitoring and accountability (PMA) survey 2019 dataset. PLOS Digit Heal. 2023;2(10):e0000345. 10.1371/journal.pdig.0000345 . 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 Inf Decis Mak. 2023;23(1):1–17. 10.1186/s12911-023-02102-w . 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(1). 10.1186/s12913-024-11932-x . Kassie SY, et al. Predictors of community-based health insurance enrollment among reproductive-age women in Ethiopia based on the EDHS 2019 dataset: a study using SHAP analysis technique, 2024. Front Public Heal. 2025;13:1–12. 10.3389/fpubh.2025.1448055 . Sarnak D, Anglewicz P, Ahmed S. Unmet need and intention to use as predictors of adoption of contraception in 10 Performance Monitoring for Action geographies. SSM - Popul Heal. 2023;22:101365. 10.1016/j.ssmph.2023.101365 . Belachew TB, Asmamaw DB, Belachew AB, Bayafers MW, Negash WD. Married women’s decision-making autonomy on modern contraceptive use and its associated factors in high fertile sub-Saharan African countries: a multi-level analysis of Demographic and Health Surveys. Arch Public Heal. 2023;81(1):1–9. 10.1186/s13690-023-01210-3 . Fantaye FT, Sene KM, Damtew SA. Contraceptive use intention and determinants among nonuser reproductive-aged married women in Ethiopia based on the theory of planned behavior. Discov public Heal. 2025;22(1). 10.1186/s12982-025-00723-2 . Additional Declarations No competing interests reported. Supplementary Files SupplementalFigure1.png SupplementalFigure1.png SupplementalFigure2.png SupplementalFigure3.png SupplementalFigure4.png Cite Share Download PDF Status: Under Review Version 1 posted Reviewers invited by journal 14 Nov, 2025 Editor invited by journal 21 Oct, 2025 Editor assigned by journal 27 Aug, 2025 Submission checks completed at journal 27 Aug, 2025 First submitted to journal 25 Aug, 2025 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-7454492","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":550111422,"identity":"3d5663d6-6962-492a-88f5-2aaccc5569af","order_by":0,"name":"Melaku Alelign Mengstie","email":"data:image/png;base64,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","orcid":"","institution":"University of Gondar","correspondingAuthor":true,"prefix":"","firstName":"Melaku","middleName":"Alelign","lastName":"Mengstie","suffix":""},{"id":550111423,"identity":"8bea4a96-3dc4-476f-a4b5-77f8836a6c5e","order_by":1,"name":"Alexander Takele Mengesha","email":"","orcid":"","institution":"University of Gondar","correspondingAuthor":false,"prefix":"","firstName":"Alexander","middleName":"Takele","lastName":"Mengesha","suffix":""}],"badges":[],"createdAt":"2025-08-25 13:53:39","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7454492/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7454492/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":96917558,"identity":"0f7172fa-4165-42d1-9c0c-3671f449aeb1","added_by":"auto","created_at":"2025-11-27 14:10:07","extension":"jpeg","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":115106,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/0c9b24fa99a91eb42fd0d1bc.jpeg"},{"id":96918569,"identity":"54e98a05-93aa-4693-881c-1fa7d32ccfa0","added_by":"auto","created_at":"2025-11-27 14:12:08","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":80579,"visible":true,"origin":"","legend":"","description":"","filename":"RevisedUnMetNeedManuscriptFinalVersionV3.docx","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/9615b27fcc1161b9ec5c59ae.docx"},{"id":96822501,"identity":"154f5347-2fd5-482d-9a7f-8f38401bb9e3","added_by":"auto","created_at":"2025-11-26 12:21:11","extension":"jpeg","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":110775,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/3b03052a6717b320fd1ecb0b.jpeg"},{"id":96822505,"identity":"608c995e-b70d-4ea9-a2a4-eb163f9992a1","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"jpeg","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96636,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/ac3af179a23646a822f6b99f.jpeg"},{"id":96822507,"identity":"b6102172-5d80-42f4-9ae1-abfdd4151549","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"json","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5468,"visible":true,"origin":"","legend":"","description":"","filename":"5a8babd7e5884e9bb1ec4a5058d83546.json","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/3787a54b19a3fe598eee6d30.json"},{"id":96822512,"identity":"fb9a5aaf-b4ad-439c-ab06-67c2922e32d6","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":20807,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/81e48074dcdace14e8268eb6.png"},{"id":96822509,"identity":"7998eb93-3189-47f2-ac00-fa5ad1fc32dd","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":51517,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/2873c2fd7e8b125de96c2ffa.png"},{"id":96917016,"identity":"8582fb12-59c5-4b81-91ae-846f6eaf9075","added_by":"auto","created_at":"2025-11-27 14:09:09","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":63437,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure3.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/fc5a4ac00fb1f6f4e04b5a4d.png"},{"id":96917768,"identity":"004fab37-e4bf-4663-a7d6-ccf5a3310685","added_by":"auto","created_at":"2025-11-27 14:10:32","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":108900,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure4.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/5280840da50fdaebae831813.png"},{"id":96822519,"identity":"0697da93-e5f5-4a5b-ad3d-c9148b4f941e","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"xml","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":83437,"visible":true,"origin":"","legend":"","description":"","filename":"5a8babd7e5884e9bb1ec4a5058d835461enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/d8ff25b85aa4f8f82833216b.xml"},{"id":96822517,"identity":"0861e01c-2435-4c4b-a9f5-150f000522b7","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"jpeg","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":115106,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/ad4e56e7f0f0a818b54436ef.jpeg"},{"id":96822514,"identity":"ffa1c09c-7a79-4127-88c1-21cdbc151bb8","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"jpeg","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":110775,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/96d98bc000dcdf1a0a4def38.jpeg"},{"id":96917620,"identity":"986cdd2a-2823-42c6-90f4-9d9fefc98536","added_by":"auto","created_at":"2025-11-27 14:10:14","extension":"jpeg","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":96636,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/906172bebd9f516acb5ee485.jpeg"},{"id":96918610,"identity":"27f40712-298f-4d17-8c8b-449e27d97822","added_by":"auto","created_at":"2025-11-27 14:12:12","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32022,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/738274acaef5e4f95c5d9a39.png"},{"id":96822515,"identity":"4f61b8df-68dd-4127-a786-cacd5f82b22a","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29739,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/b648c152db8c9e0a51988df8.png"},{"id":96822518,"identity":"4b104a1b-d430-4117-9f12-de73fbc810fb","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25418,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure3.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/5396d0071491c555d8ce816b.png"},{"id":96822521,"identity":"ff565b12-ff03-4bfe-a67e-7be297f60352","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"xml","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":82711,"visible":true,"origin":"","legend":"","description":"","filename":"5a8babd7e5884e9bb1ec4a5058d835461structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/0754f731be03cce0b16313e3.xml"},{"id":96917732,"identity":"b4a3a1a8-880d-4a8b-aaa3-d17cd1827fff","added_by":"auto","created_at":"2025-11-27 14:10:27","extension":"html","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":92685,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/738e7ea27e25cac5dbc439c6.html"},{"id":96822497,"identity":"a9644623-88a3-4931-b51c-0292ac2e5aa1","added_by":"auto","created_at":"2025-11-26 12:21:11","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":115106,"visible":true,"origin":"","legend":"\u003cp\u003eFramework for the development of the predictive model identifying unmet contraceptive needs.\u003c/p\u003e","description":"","filename":"Figure1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/f257ba65f15a5b5bd10c5f11.jpeg"},{"id":96822503,"identity":"f849a717-b585-4f8c-8936-795d0bb96689","added_by":"auto","created_at":"2025-11-26 12:21:12","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":110775,"visible":true,"origin":"","legend":"\u003cp\u003eInterpretation of potential risk factors for unmet contraceptive needs identified by the model\u003c/p\u003e","description":"","filename":"Figure2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/260b19379990a22975f37328.jpeg"},{"id":96918055,"identity":"6e0f5c75-31e9-4d11-a90f-b5547e08fb56","added_by":"auto","created_at":"2025-11-27 14:11:05","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":96636,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP waterfall plot illustrating the contribution of individual features to the prediction for a single case.\u003c/p\u003e","description":"","filename":"Figure3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/32caaac6ca325e9a8fe91fd7.jpeg"},{"id":96923031,"identity":"e66450a9-b513-4c3a-9b15-8197f162e7d0","added_by":"auto","created_at":"2025-11-27 14:20:38","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1222254,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/53dec854-bb3e-48cb-b503-c7679d4d7970.pdf"},{"id":96822499,"identity":"aaa3eb3d-2d25-4c30-954e-58dc6aea2dea","added_by":"auto","created_at":"2025-11-26 12:21:11","extension":"png","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":20807,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/a9719945ce6fd09d26e06f7e.png"},{"id":96822496,"identity":"3b9ac838-c088-4fde-8d13-690c4cbe08dc","added_by":"auto","created_at":"2025-11-26 12:21:11","extension":"png","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":20807,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/216f96a11808911625ddf7af.png"},{"id":96822500,"identity":"811aac4e-47d8-4a4d-b5ad-0c6e3148c200","added_by":"auto","created_at":"2025-11-26 12:21:11","extension":"png","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":51517,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/30ecb07b39b99d93d5744cf0.png"},{"id":96917164,"identity":"6fcb2da9-49f9-47e4-a2fd-b1d95d1418d3","added_by":"auto","created_at":"2025-11-27 14:09:19","extension":"png","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":63437,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure3.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/0ef0c5294f77e3e45e5be428.png"},{"id":96918052,"identity":"0fc2f6cc-1a04-4953-9800-4d3f5fdecb57","added_by":"auto","created_at":"2025-11-27 14:11:05","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":108900,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementalFigure4.png","url":"https://assets-eu.researchsquare.com/files/rs-7454492/v1/ab32395f40bdc40c3f03bfef.png"}],"financialInterests":"No competing interests reported.","formattedTitle":"Ensemble Machine Learning with SHAP Interpretability for Predicting Unmet Contraceptive Needs in Ethiopia","fulltext":[{"header":"Key Messages","content":"\u003cp\u003e\u003cstrong\u003eWhat is already known about this topic\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUnmet contraceptive needs persist in Ethiopia, and previous predictive models lacked accuracy and interpretability.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWhat this study adds\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study applies an ensemble machine learning approach enhanced with SHAP to predict unmet contraceptive needs, identifying key factors like contraceptive awareness and fertility preferences.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHow might this study affect research, practice, or policy\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe model’s transparency and high accuracy can guide targeted reproductive health interventions and support data-driven policy aligned with SDG 3.7.\u003c/p\u003e"},{"header":"1. INTRODUCTION","content":"\u003cp\u003eFamily planning plays a crucial role in public health by enabling individuals to control the timing and number of children. This improves maternal and child health, reduces unintended pregnancies, and supports women's education and careers. However, unmet contraceptive needs remain a pressing issue, particularly in developing regions[1]. Unmet contraceptive needs when individuals who wish to delay or prevent pregnancy are not using any method of contraception pose a persistent public health challenge[2].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOver 218 million women globally face an unmet need for family planning, contributing to a high rate of unintended pregnancies due to limited use of modern contraceptive methods[3]. In sub-Saharan Africa, unintended pregnancies remain widespread, influenced by factors such as education, contraceptive awareness, marital status, and regional disparities[4].\u0026nbsp;Unintended pregnancies lead to serious health risks, including preterm labour, unsafe abortions, and maternal depression[5][6].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn Ethiopia, A significant proportion of women express a desire to space or limit births but still lack access to appropriate contraceptive methods. National surveys have reported high rates of unintended pregnancies and persistent regional and socioeconomic disparities in contraceptive use[7]. Despite progress in long-acting and permanent contraceptive use, rising from 17.9% in 2014 to 35.5% in 2019, the contraceptive prevalence rate is still far below the 2020 target of 55%[8].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEthiopia’s Federal Ministry of Health aims to reduce the unmet need for family planning from 22% to 17% [9] and increase the contraceptive prevalence rate to 54% by 2030[10][11]. Data-driven approaches are vital for effectively identifying women at risk of unmet contraceptive needs. Despite efforts, many face barriers to access, partly due to a lack of accurate predictive tools. Ensemble machine learning offers precise risk prediction, helping policymakers and healthcare providers target interventions better. Improving contraceptive prevalence and reducing unmet need align with 2030 Sustainable Development Goal 3.7 to ensure universal access to sexual and reproductive health services[12].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMachine Learning Algorithms\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMachine Learning (ML) algorithms learn patterns from socio-demographic, behavioral, and health data to capture complex, non-linear relationships often missed by traditional statistics[13]. Ensemble methods, Random Forest, CatBoost, LightGBM, and XGBoost combine multiple weak learners to enhance accuracy and robustness[14]; CatBoost excels with categorical data and overfitting prevention[15]; LightGBM accelerates training with leaf-wise splits[16] ; XGBoost uses gradient optimization and regularization[17]. and Random Forest improves generalization by aggregating many decision trees[18].\u003c/p\u003e\n\u003cp\u003eIn family planning, ML predicts unmet contraceptive needs, evaluates programs, identifies access barriers, and forecasts demand. When integrated with mobile health platforms, ML empowers personalized reproductive health decisions[19][20]. Explainable AI (XAI) techniques like SHAP enhance transparency by clarifying feature influence, enabling data-driven, equitable policy interventions. Deploying ML in Ethiopia could optimize family planning resources, improve service delivery, and help reach a 54% contraceptive prevalence rate by 2030, reducing unintended pregnancies and improving maternal health.\u003c/p\u003e\n\u003cp\u003eThis study develops an ensemble machine-learning model to predict unmet contraceptive needs among Ethiopian reproductive-age women. By integrating multiple algorithms, it aims to enhance prediction accuracy and identify key factors that drive unmet needs. The findings will inform policies and interventions aimed at enhancing family planning services in Ethiopia.\u003c/p\u003e\n\u003cp\u003eThe following questions guide the research: What are the relevant features for constructing the predictive model for unmet contraceptive needs?Which ensemble machine learning algorithm is most suitable for predicting unmet contraceptive needs among Ethiopian reproductive-age women? How can Explainable AI techniques be applied to enhance the interpretability of the predictive model, and in identifying the most influential factors influencing unmet contraceptive needs?\u003c/p\u003e"},{"header":"II.\tRelated Works ","content":"\u003cp\u003eD. Mamo et.al [21]predicted unintended pregnancy using six machine learning algorithms on 7,193 secondary data points from the 2016 EDHS survey. Extra Trees emerged as the top model, achieving 92.8% AUC with default hyperparameter tuning, identifying predictors such as region, ideal number of children, and education. However, ensemble methods were not utilized, prompting our focus on advanced ensemble algorithms and XAI for unmet contraceptive needs in Ethiopia.\u0026nbsp;S. D. Kebede et.al\u0026nbsp;[22]used machine learning to predict the unmet contraceptive needs of 5,819 women from the 2019 PMA survey, identifying Random Forest as the best model (85% accuracy, 0.93 AUC). SHAP revealed key predictors like partner disapproval and regional factors, but ensemble techniques were omitted, highlighting the need for our advanced approach to integrating XAI.\u0026nbsp;S. Kebede et al\u0026nbsp;[23]explored contraceptive discontinuation using Random Forest (68% accuracy) on EDHS data, identifying age, education, and family size as key predictors. However, ensemble techniques and XAI were absent, necessitating our study’s innovative approach to improve predictions and transparency. Similarly, T. Yehuala\u0026nbsp;[24]\u0026nbsp;applied XGBoost to predict non-use of modern family planning among 92,564 East African women, identifying education, age, rural residence, marital status, contraceptive knowledge, and smoking as key predictors. Similarly, S. \u0026nbsp;Kassie et al.\u0026nbsp;[25]\u0026nbsp;used Random Forest to predict community-based health insurance enrollment among 9,013 Ethiopian women, highlighting residence, wealth, household head age, husband’s education, media exposure, family size, and number of under-five children as influential factors. While both studies demonstrate the power of machine learning for health prediction, they lacked systematic multi-algorithm benchmarking and interpretable, policy-focused insights. Our study addresses these gaps by combining ensemble algorithms with SHAP to provide accurate and explainable predictions specific to unmet contraceptive needs.\u003c/p\u003e\n\u003cp\u003ePrevious studies [21]to[23]provided valuable insights but lacked the innovation of ensemble learning and Explainable AI (XAI). Our research addresses this gap by integrating advanced ensemble algorithms with XAI, enhancing predictive accuracy, interpretability, and policy relevance. Similarly, T. Z. Yehuala [24] and S. Y. Kassie et al. [25] applied machine learning to reproductive health outcomes, but did not focus on unmet contraceptive needs or leverage ensemble methods with XAI. Our study fills this gap, delivering more reliable and interpretable predictions of unmet contraceptive needs in Ethiopia, supporting targeted family planning interventions. Provided valuable insights but lacked the innovation of ensemble learning and Explainable AI. Our research addresses this gap by integrating advanced ensemble algorithms with XAI, enhancing predictive accuracy, interpretability, and policy relevance. Similarly,[24] and [25] \u0026nbsp; applied machine learning to reproductive health outcomes, but did not focus on unmet contraceptive needs. Our study fills this gap, delivering more reliable and interpretable predictions of unmet contraceptive needs in Ethiopia, supporting targeted family planning interventions.\u003c/p\u003e"},{"header":"III.\tMaterials and Methods ","content":"\u003cp\u003eFigure 1 illustrates the workflow from raw data to actionable insights. It covers data cleaning and transformation, balancing with SMOTETomek, selecting key features, training multiple models, and using SHAP to interpret the results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA.\u0026nbsp;\u0026nbsp;Data Collection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe researchers analyzed data from the Ethiopian Demographic and Health Surveys (2011–2016), which included socio-demographic, behavioral, and health-related information. This comprehensive dataset allowed them to explore patterns and factors influencing contraceptive access and use across different populations and regions in Ethiopia.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eB.\u0026nbsp;\u0026nbsp;Data preprocessing\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAfter collecting the secondary dataset from EDHS, the next step is data preprocessing, which includes data cleaning, transformation, feature selection, class balancing, data splitting, and other necessary steps.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1.\u0026nbsp; \u0026nbsp;Data Cleaning\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMissing numerical values such as family size, husband’s age, and child’s birth size were filled using rounded means (5, 37, 41). For important categorical variables like pregnancy intention, contraceptive decision-making, and HIV testing method, imputation maintained the original data distribution, ensuring reliable analysis of reproductive and socio-demographic factors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.\u0026nbsp; \u0026nbsp;Data Transformation\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWith guidance from domain experts, continuous variables such as family size, birth interval, antenatal visits, and BMI were grouped into meaningful categories to improve model interpretability. Key categorical features, including place of delivery, water source, and housing materials, were recoded for consistency. These expert-informed transformations aligned the data with real-world health contexts, supporting more effective modeling.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eC.\u0026nbsp;\u0026nbsp;Data Splitting\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe dataset was split into 20% test data, with the remaining 80% further divided into training and validation sets (80:20). This ensured consistent class distribution across all subsets for reliable model evaluation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.\u0026nbsp; \u0026nbsp;Feature Selection\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe used Recursive Feature Elimination to determine the optimal threshold for feature selection, which resulted in identifying 21 key features. By iteratively eliminating less important variables, RFE sets this threshold and retains only the most relevant predictors for subsequent model development, thereby enhancing the model’s efficiency and effectiveness.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.\u0026nbsp; \u0026nbsp; Addressing Class Imbalance\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo handle class imbalance in the training set, SMOTETomek was used, combining oversampling of the minority class with under-sampling through Tomek links. This balanced the ‘unmet need’ classes to 8,410 each, reducing bias toward the majority. Applied only to the training data, it preserved the validation and test sets, ensuring reliable model evaluation (see\u0026nbsp;Supplemental Figure 1).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eD.\u0026nbsp;\u0026nbsp;Model Development\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMultiple models were developed to predict unmet contraceptive needs. Ensemble methods Random Forest, CatBoost, XGBoost, and LightGBM were used alongside Logistic Regression and SVM as benchmarks, allowing comparison between more complex and simpler approaches.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eE.\u0026nbsp;\u0026nbsp;Model Validation and Evaluation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eHyperparameters were tuned, and stratified five-fold cross-validation ensured model stability. The final evaluation on the test set used accuracy, precision, recall, F1 score, and AUC-ROC. SHAP analysis was applied to interpret the models and identify key predictive features.\u003c/p\u003e"},{"header":"IV.\tResults and Discussion ","content":"\u003cp\u003e\u003cstrong\u003eA.\u0026nbsp;\u0026nbsp;Experimental Result\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnalysis of Relevant Features\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUsing Recursive Feature Elimination with a Random Forest classifier, 21 key features were selected from 88, prominent a balance between accuracy and simplicity. These features highlighted the most important demographic, reproductive, behavioral, and health-related factors. Leading predictors included contraception exposure, fertility desire, pregnancy intention, intention to use, and past family planning use. Other influential factors, such as region, wealth index, contraceptive decision-making, HIV testing, and partner’s education, further improved model performance and interpretability (see Supplemental Figure 2).Top of FormBottom of Form\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePredictive Model Development\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTrained on 21 key features selected via RFE, the models captured complex patterns across demographic, socioeconomic, healthcare, and reproductive domains. Performance metrics from stratified five-fold cross-validation are shown in Table 1, with final test results in Table 2.\u003c/p\u003e\n\u003cp\u003eTable 1 Five-Fold Cross-Validation Performance\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"539\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1 Score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e97.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e94.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.28\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCatBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.52\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e97.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e95.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.69\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e97.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e95.99\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.53\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLightGBM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e95.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e96.52\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLogistic Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e86.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e86.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e86.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e86.36\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSupport Vector Machine\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e93.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e93.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e93.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e93.21\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eTable 2 Performance of Selected Ensemble Algorithms using Normal Test Set\u0026nbsp;\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"548\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1 Score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e95.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e93.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e87.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e90.63\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCatBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e95.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e91.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e87.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e89.87\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e95.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e93.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e88.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e90.90\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLightGBM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e95.21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e91.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e88.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e90.32\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLogistic Regression\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e83.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e63.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e80.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e71.37\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSupport Vector Machine\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e92.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e84.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e85.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e85.35\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003eBest Model Performance and Selection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eXGBoost proved to be the top model for predicting unmet contraceptive needs, achieving 95.55% accuracy, 93.67% precision, 88.29% recall, and a 90.90% F1 score on the test set. It consistently outperformed other models in cross-validation by balancing sensitivity and reducing false positives. As shown in Supplemental Figure 3, ROC analysis indicated that XGBoost, Random Forest, and LightGBM reached near-perfect AUCs of 0.99, demonstrating the superior predictive power of tree-based ensembles compared to traditional methods like Logistic Regression and SVM. These findings establish XGBoost as the most reliable and effective model for this task.\u003c/p\u003e\n\u003ch3\u003eDiscussion\u003c/h3\u003e\n\u003cp\u003e\u003cstrong\u003e1.\u0026nbsp; \u0026nbsp;What are the relevant features for constructing the predictive model and the potential risk factors for unmet contraceptive needs?\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAs presented in Supplemental Figure 2, the top 21 features selected through Recursive Feature Elimination represent the key risk factors associated with unmet contraceptive needs in Ethiopia. These features highlight the critical role of exposure to contraceptive information, previous use of family planning methods, pregnancy intention, intention to use contraception, and current pregnancy status. Fertility preferences, including the desire for more children and the ideal number of children, along with awareness of alternative family planning methods and decision-making autonomy, also emerge as important predictors. Together, these variables illustrate the complex relationships among reproductive intentions, access to information, personal agency, and healthcare engagement. This selection of features provides valuable guidance for developing targeted, data-driven reproductive health interventions aimed at reducing disparities and improving outcomes. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBottom of Form\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.\u0026nbsp; \u0026nbsp;Which ensemble machine-learning technique is most effective for building a predictive model for unmet contraceptive needs among Ethiopian women?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eXGBoost emerged as the most effective ensemble model, utilizing 21 key features identified through Recursive Feature Elimination. During five-fold cross-validation, it achieved a top precision of 97.59%, alongside strong recall and F1 scores, accurately identifying cases while minimizing false positives. On the test set, XGBoost outperformed all other models with 95.55% accuracy, 93.67% precision, 88.29% recall, and a 90.90% F1 score, demonstrating robustness and generalizability.\u003c/p\u003e\n\u003cp\u003eIts strength lies in capturing complex, nonlinear patterns, handling missing data, and preventing overfitting through regularization critical for modeling Ethiopia’s diverse socioeconomic and reproductive factors. Beyond impressive performance metrics, XGBoost provides actionable insights, making it a powerful tool for policymakers and health professionals addressing unmet contraceptive needs.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBottom of Form\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.\u0026nbsp; \u0026nbsp;How can Explainable AI techniques be applied to enhance the interpretability of the predictive model, and in identifying the most influential factors influencing unmet contraceptive needs?\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eExplainable AI techniques, such as SHAP, improve the transparency and interpretability of predictive models for unmet contraceptive needs. Using SHAP, this study provided both global and local explanations of the XGBoost model’s predictions.\u003c/p\u003e\n\u003cp\u003eFigure 2 shows the average absolute SHAP values, highlighting the top eight influential features: exposure to contraception information, prior family planning use, desire for more children, intention to use contraception, pregnancy intention, current pregnancy status, knowledge of other family planning methods, and primary decision-maker on contraception. These factors strongly shape the model’s outcomes.\u003c/p\u003e\n\u003cp\u003eAs presented,\u0026nbsp;Supplemental Figure\u0026nbsp;4 presents the SHAP summary plot illustrating both the magnitude and direction of each feature’s effect on unmet contraceptive needs, showing which features increase risk (e.g., desire for more children, pregnancy intention) and which decrease risk (e.g., exposure to contraception information, prior family planning use).\u003c/p\u003e\n\u003cp\u003eAt the individual level, SHAP waterfall plots (Figure 3) break down how features push predictions toward higher or lower risk, revealing personalized influences that may differ from global trends. This granular insight fosters ethical, informed public health decisions by explaining not just what the model predicts, but why.\u003c/p\u003e\n\u003cp\u003eOur study confirms key risk factors for unmet contraceptive needs in Ethiopia, with exposure to contraceptive information as a critical determinant[2]. \u0026nbsp;Consistent with prior research, lack of media exposure increases unmet need odds by 1.32 times in high-fertility areas. These results highlight improving access to reproductive health information as a vital strategy to reduce unmet contraceptive needs.\u003c/p\u003e\n\u003cp\u003eResearch in East Africa shows that a woman’s intention to use contraception strongly predicts uptake, with intention within a year linked to higher adoption rates[26]. \u0026nbsp;In Ethiopia, greater decision-making autonomy reduces unmet contraceptive need, while unintended or current pregnancies are consistently associated with higher unmet need[27]. Additionally, women desiring more children are significantly more likely to report unmet need, highlighting the influence of fertility preferences on contraceptive use[28].\u003c/p\u003e\n\u003cp\u003eKey findings reveal that unmet contraceptive needs in Ethiopia are mainly driven by limited family planning information, low decision-making autonomy, and unclear fertility intentions. Women lacking exposure to contraceptive messaging and those with a strong desire for more children face higher risks. SHAP analysis confirms these factors as the most influential, emphasizing the critical role of empowerment and informed reproductive choices.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLimitations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study, while insightful, has limitations. It relies on cross-sectional EDHS data, restricting causal and temporal analysis. Self-reported data may introduce recall bias, impacting accuracy. SHAP improves interpretability but may miss complex feature interactions and does not imply causation. Lastly, the model’s applicability outside Ethiopia is uncertain due to differing sociocultural and healthcare contexts. Future work should use longitudinal data and validate the model in diverse populations for greater robustness and policy relevance.\u003c/p\u003e"},{"header":"V.\tConclusion and Recommendation ","content":"\u003cp\u003e\u003cstrong\u003eConclusion\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUnmet contraceptive needs remain a major challenge in Ethiopia, driven by limited access to family planning information, low women’s autonomy, and unclear fertility intentions. These factors hinder effective contraceptive use and impact reproductive health outcomes. Using EDHS data, our study applied XGBoost and Explainable AI to predict unmet needs with high accuracy. Key predictors identified include pregnancy intention, prior contraceptive use, desire for more children, exposure to family planning information, and decision-making autonomy. SHAP analysis clarified the influence of these factors, enhancing model transparency.\u003c/p\u003e\n\u003cp\u003eThese findings underscore the urgent need for policies that expand access to reproductive health information and empower women’s decision-making. Integrating AI-driven predictive tools into health systems can enable targeted interventions, improving resource allocation and program effectiveness. Tailored counseling aligned with women’s fertility goals and broader efforts to promote gender equity are essential to reducing unmet contraceptive needs. Our work provides actionable insights to guide policymakers and health professionals in advancing reproductive health equity and achieving national and global family planning goals.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eContribution\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe research advances public health and AI integration by developing a robust predictive framework employing ensemble algorithms (XGBoost, CatBoost, LightGBM, Random Forest), balanced via SMOTETomek and optimized through Recursive Feature Elimination. SHAP explainability ensures transparent outputs, empowering policymakers to focus on critical factors like contraceptive knowledge and women’s autonomy. Leveraging publicly available data, the scalable model offers a practical tool adaptable to similar low-resource settings for data-driven reproductive health strategies.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRecommendation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBased on our findings, we recommend integrating XGBoost with SHAP into national health planning to identify high-risk groups and improve resource allocation. Expanding targeted health education through community outreach, empowering women’s reproductive autonomy, and enhancing provider counseling are essential. Future research should use longitudinal and real-time data and test this framework in other sub-Saharan contexts. Successful implementation requires strong digital infrastructure, institutional support, and capacity building for sustainable, data-driven public health decisions.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics Approval and Consent to Participate\u003c/strong\u003e\u003cp\u003eNot applicable.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003ch2\u003eClinical Trial Number\u003c/h2\u003e\u003cp\u003eNot applicable.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003ch2\u003eConsent for Publication\u003c/h2\u003e\u003cp\u003eNot applicable.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003ch2\u003eCompeting Interests\u003c/h2\u003e\u003cp\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eNo funding was received for this research.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAuthors' Contributions\u0026bull;Melaku Alelign Mengstie: Conceptualization, data analysis, and manuscript writing.\u0026bull;Alexander Takele Mengesha: data collection, methodology, and Interpretation.All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eWe sincerely thank the Ethiopian Demographic and Health Surveys for providing the essential data for this study. We also gratefully acknowledge the public health and reproductive health experts from the University of Gondar, whose guidance was vital in data processing, feature selection, and interpreting results, ensuring the study\u0026rsquo;s relevance to real-world challenges and policy\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe dataset used in this study, the Ethiopian Demographic and Health Survey (EDHS), is publicly available. All data and materials related to the study can be accessed upon reasonable request from the corresponding author.The dataset is available at the EDHS website: [https://dhsprogram.com/data/available-datasets.cfm](https:/dhsprogram.com/data/available-datasets.cfm)\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSmith M, Keckley P. Adding it up. Hosp Heal Networks. 2017;91(10):64\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/0162643418759341\u003c/span\u003e\u003cspan address=\"10.1177/0162643418759341\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAsmamaw DB, Negash WD. Magnitude of unmet need for family planning and its predictors among reproductive age women in high fertility regions of Ethiopia: Evidence from Ethiopian Demographic and Health Survey. BMC Womens Health. 2022;22(1):1\u0026ndash;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12905-022-01982-w\u003c/span\u003e\u003cspan address=\"10.1186/s12905-022-01982-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNemser B, Addofoh N. Contextual factors associated with contraceptive utilization and unmet need among sexually active unmarried women in Kenya: A multilevel regression analysis, \u003cem\u003ePLoS One\u003c/em\u003e, vol. 17, no. 6 June, pp. 1\u0026ndash;21, 2022, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0270516\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0270516\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAyalew HG, et al. Prevalence and factors associated with unintended pregnancy among adolescent girls and young women in sub-Saharan Africa, a multilevel analysis. BMC Womens Health. 2022;22(1):1\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12905-022-02048-7\u003c/span\u003e\u003cspan address=\"10.1186/s12905-022-02048-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGelaw KA, Atalay YA, Gebeyehu NA. Unintended pregnancy and contraceptive use among women in low \u0026ndash; and middle \u0026ndash; income countries: systematic review and meta \u0026ndash; analysis, pp. 1\u0026ndash;13, 2023, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40834-023-00255-7\u003c/span\u003e\u003cspan address=\"10.1186/s40834-023-00255-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGaro MG, Abe SG, Girsha WD, Daka DW. Unmet need for family planning and associated factors among currently married women of reproductive age in Bishoftu town, Eastern Ethiopia. PLoS ONE. 2021;16(12):1\u0026ndash;16. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pone.0260972\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0260972\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEthiopian Public Health Institute (EPHI) and ICF, Ethiopia Mini Demographic and Health Survey 2019: Final Report. 2021. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://dhsprogram.com/pubs/pdf/FR363/FR363.pdf\u003c/span\u003e\u003cspan address=\"https://dhsprogram.com/pubs/pdf/FR363/FR363.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEthiopian Ministry of Health. Health Sector Transformation Plan II 2020/2021\u0026ndash;2024/2025, \u003cem\u003eEthiop. Minist. Heal.\u003c/em\u003e, vol. 25, no. February, pp. 1\u0026ndash;128, 2021.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVision P. Ethiopia \u0026rsquo; s FP2030 Commitment Approach to Accountability, no. 7200, 2025.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eUNFPA Ethiopia. | Ethiopia launches a costed family planning implementation plan. Accessed: Apr. 03, 2025. [Online]. Available: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ethiopia.unfpa.org/en/news/ethiopia-launches-costed-family-planning-implementation-plan\u003c/span\u003e\u003cspan address=\"https://ethiopia.unfpa.org/en/news/ethiopia-launches-costed-family-planning-implementation-plan\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShumet T, Geda NR, Hassan JA. Barriers to modern contraceptive utilization in Ethiopia. Contracept Reprod Med. 2024;9(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40834-024-00311-w\u003c/span\u003e\u003cspan address=\"10.1186/s40834-024-00311-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNations U. World Contraceptive Use, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.un.org/development/desa/pd/data/world-contraceptive-use\u003c/span\u003e\u003cspan address=\"https://www.un.org/development/desa/pd/data/world-contraceptive-use\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTaye SBM, et al. Predicting Unmet Contraceptive Needs Using Machine Learning Techniques. J Health Inf Developing Ctries. 2021;15(1):1\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSeries C. Extreme gradient boosting (XGBoost) method in making forecasting application and analysis of USD exchange rates against rupiah Extreme gradient boosting (XGBoost) method in making forecasting application and analysis of USD exchange rates against rupi, pp. 0\u0026ndash;11, 2021, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1088/1742-6596/1722/1/012016\u003c/span\u003e\u003cspan address=\"10.1088/1742-6596/1722/1/012016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTanha J, Abdi Y, Samadi N, Razzaghi N, Asadpour M. Boosting methods for multi-class imbalanced data classification: an experimental review. J Big Data. 2020;7(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40537-020-00349-y\u003c/span\u003e\u003cspan address=\"10.1186/s40537-020-00349-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSyam P, Chand S, Divya G. A Light Gradient Boosting Machine Regression Model for Prediction of Agriculture Insurance Cost over Linear Regression, pp. 200\u0026ndash;8, 2022, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3233/APC220027\u003c/span\u003e\u003cspan address=\"10.3233/APC220027\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGanaie MA, Hu M, Tanveer* M, Suganthan* PN. Ensemble deep learning: A review, 2021.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSmirani LK, Yamani HA, Menzli LJ, Boulahia JA. Using Ensemble Learning Algorithms to Predict Student Failure and Enabling Customized Educational Paths, \u003cem\u003eSci. Program.\u003c/em\u003e, vol. 2022, 2022. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1155/2022/3805235\u003c/span\u003e\u003cspan address=\"10.1155/2022/3805235\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eObermeyer EJ, Z., Emanuel. Predicting the Future\u0026mdash;Big Data, Machine Learning, and Clinical Medicine. N Engl J Med. 2016;375(13):1216\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1056/NEJMp1606181\u003c/span\u003e\u003cspan address=\"10.1056/NEJMp1606181\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eClassification HI. A Comprehensive Review on Machine Learning in Healthcare Industry: Classification, Restrictions, Opportunities and Challenges, 2023.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMamo DN, Gebremariam YH, Adem JB, Kebede SD, Walle AD. Machine learning to predict unintended pregnancy among reproductive-age women in Ethiopia: evidence from EDHS 2016. BMC Womens Health. 2024;24(1):1\u0026ndash;11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12905-024-02893-8\u003c/span\u003e\u003cspan address=\"10.1186/s12905-024-02893-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKebede SD, Mamo DN, Adem JB, Semagn BE, Walle AD. Machine learning modeling for identifying predictors of unmet need for family planning among married/in-union women in Ethiopia: Evidence from performance monitoring and accountability (PMA) survey 2019 dataset. PLOS Digit Heal. 2023;2(10):e0000345. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1371/journal.pdig.0000345\u003c/span\u003e\u003cspan address=\"10.1371/journal.pdig.0000345\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\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 Inf Decis Mak. 2023;23(1):1\u0026ndash;17. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12911-023-02102-w\u003c/span\u003e\u003cspan address=\"10.1186/s12911-023-02102-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\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(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12913-024-11932-x\u003c/span\u003e\u003cspan address=\"10.1186/s12913-024-11932-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKassie SY, et al. Predictors of community-based health insurance enrollment among reproductive-age women in Ethiopia based on the EDHS 2019 dataset: a study using SHAP analysis technique, 2024. Front Public Heal. 2025;13:1\u0026ndash;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fpubh.2025.1448055\u003c/span\u003e\u003cspan address=\"10.3389/fpubh.2025.1448055\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSarnak D, Anglewicz P, Ahmed S. Unmet need and intention to use as predictors of adoption of contraception in 10 Performance Monitoring for Action geographies. SSM - Popul Heal. 2023;22:101365. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ssmph.2023.101365\u003c/span\u003e\u003cspan address=\"10.1016/j.ssmph.2023.101365\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBelachew TB, Asmamaw DB, Belachew AB, Bayafers MW, Negash WD. Married women\u0026rsquo;s decision-making autonomy on modern contraceptive use and its associated factors in high fertile sub-Saharan African countries: a multi-level analysis of Demographic and Health Surveys. Arch Public Heal. 2023;81(1):1\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13690-023-01210-3\u003c/span\u003e\u003cspan address=\"10.1186/s13690-023-01210-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFantaye FT, Sene KM, Damtew SA. Contraceptive use intention and determinants among nonuser reproductive-aged married women in Ethiopia based on the theory of planned behavior. Discov public Heal. 2025;22(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12982-025-00723-2\u003c/span\u003e\u003cspan address=\"10.1186/s12982-025-00723-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\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-health-services-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bhsr","sideBox":"Learn more about [BMC Health Services Research](http://bmchealthservres.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/BHSR/default.aspx","title":"BMC Health Services Research","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Ensemble learning, Explainable AI, Family planning, Machine learning, Reproductive health, SHAP values, Unmet contraceptive needs, XGBoost","lastPublishedDoi":"10.21203/rs.3.rs-7454492/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7454492/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eUnmet contraceptive needs remain a critical challenge in global reproductive health, especially in developing countries like Ethiopia, where access to family planning is limited. Women who wish to delay or avoid pregnancy but do not use contraception face increased risks of unintended pregnancies, unsafe abortions, and adverse maternal outcomes. This study aims to develop and evaluate an ensemble machine learning model, enhanced with Explainable AI techniques, to accurately identify women at risk of unmet contraceptive needs, thereby supporting informed and transparent decision-making.\u003c/p\u003e\u003cp\u003eData from the 2011 and 2016 Ethiopian Demographic and Health Surveys were used. Ensemble models, including Random Forest, Categorical Boosting, Extreme Gradient Boosting, and Light Gradient Boosting Machine, were trained on 21 key features selected through Recursive Feature Elimination. A hybrid SMOTE-Tomek sampling technique addressed class imbalance. Stratified train-validation-test splits ensured robust performance evaluation.\u003c/p\u003e\u003cp\u003eExtreme Gradient Boosting emerged as the best-performing model, achieving 96.56% accuracy, 97.59% precision, 95.99% recall, and a 96.53% F1-score in cross-validation. On the test set, it maintained strong results with 95.55% accuracy and a 90.90% F1-score, outperforming Logistic Regression and Support Vector Machine. Receiver Operating Characteristic curve analysis confirmed its excellent classification (AUC\u0026thinsp;=\u0026thinsp;0.99). SHapley Additive exPlanations analysis highlighted key predictors driving Extreme Gradient Boosting\u0026rsquo;s predictions, including contraceptive information exposure, prior family planning use, pregnancy intention, decision-making autonomy, and fertility preferences.\u003c/p\u003e\u003cp\u003eThe findings provide interpretable, data-driven insights for targeted reproductive health interventions. Integrating such predictive models into real-time health systems may enhance family planning strategies and help achieve Ethiopia\u0026rsquo;s 2030 health goals.\u003c/p\u003e","manuscriptTitle":"Ensemble Machine Learning with SHAP Interpretability for Predicting Unmet Contraceptive Needs in Ethiopia","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-11-26 12:21:07","doi":"10.21203/rs.3.rs-7454492/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewersInvited","content":"","date":"2025-11-14T17:34:45+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-10-21T23:29:25+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-27T06:54:11+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-08-27T06:53:15+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Health Services Research","date":"2025-08-25T13:40:46+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-health-services-research","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bhsr","sideBox":"Learn more about [BMC Health Services Research](http://bmchealthservres.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/BHSR/default.aspx","title":"BMC Health Services Research","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"bec60e1a-f34f-42c3-966d-cabbbc61ae78","owner":[],"postedDate":"November 26th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-11-26T12:21:07+00:00","versionOfRecord":[],"versionCreatedAt":"2025-11-26 12:21:07","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7454492","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7454492","identity":"rs-7454492","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}