Exploring Machine Learning Insights into Long-Acting Reversible Family Planning Usage in Ethiopia: Analysis of the PMA (2021-2023) Dataset

Exploring Machine Learning Insights into Long-Acting Reversible Family Planning Usage in Ethiopia: Analysis of the PMA (2021-2023) Dataset | 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 Exploring Machine Learning Insights into Long-Acting Reversible Family Planning Usage in Ethiopia: Analysis of the PMA (2021-2023) Dataset Abraham Keffale Mengistu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6417320/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 10 Jan, 2026 Read the published version in BMC Public Health → Version 1 posted 13 You are reading this latest preprint version Abstract Introduction : Ethiopia faces challenges in Long-Acting Reversible Family Planning (LARFP) adoption despite its efficacy. Traditional statistical methods have a limited capacity to capture nonlinear determinants. This study leverages machine learning (ML) to identify predictors of LARFP use using the 2021-2023 PMA Ethiopia dataset. Methods : A nationally representative sample of 9,763 women aged 15–49 was analyzed. Twenty-four variables across geographic, socioeconomic, healthcare access, and behavioral domains were preprocessed (handling missing values, encoding, and normalization). Seven ML models (Decision Tree, XGBoost, Random Forest, Logistic Regression, SVM, KNN, Naive Bayes) were trained and evaluated via stratified 5-fold cross-validation. Performance metrics included accuracy, precision, recall, F1 score, and AUC-ROC. Results : Decision Tree outperformed other models (accuracy: 99.45%, F1: 99.55%), identifying method duration (importance=0.35), provider type (0.25), and region (0.15) as top predictors. Regional disparities were stark (SNNP: 30.59% LARFP use vs. Amhara: 15.58%). Key reasons for method choice included fewer side effects (32.3%) and long duration (15.5%). Conclusion : Tree-based ML models effectively captured complex determinants of LARFP use. Targeted interventions addressing regional disparities, provider training, and client-centered care (e.g., reducing side effects) are critical for improving uptake. Machine learning long-acting reversible contraception family planning Ethiopia PMA survey decision tree health equity Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Ethiopia has made significant strides over the past decades in attaining better reproductive health outcomes, yet obstacles to universal access to family planning services remain ( 1 , 2 ). Family planning is one of the pillars of reproductive health that enables individuals and couples to plan and have the number of children they desire and the space and time for their births( 3 ). Among the variety of family planning methods, long-acting reversible family planning (LARFP) methods such as intrauterine devices (IUDs) and implants are extremely effective in averting unintended pregnancy and maternal and child mortality( 4 – 6 ). These methods are particularly advantageous in low-resource contexts like Ethiopia, where health infrastructure and access to services are generally limited ( 5 , 6 ). Despite their effectiveness, the utilization of LARFPs in Ethiopia remains low, at only 22% among reproductive-age women ( 7 , 8 ). The low utilization is attributed to a combination of socioeconomic, cultural, and access-to-healthcare barriers, which have hindered the widespread application of these life-saving interventions ( 9 ). The Ethiopian government, in collaboration with international partners, has implemented various programs for the promotion of family planning, including the dissemination of LARFP methods( 10 , 11 ). However, the gap between their availability and use underscores the need to better understand their determinants of adoption ( 11 ). Traditional statistical approaches, such as logistic regression, have been widely used to identify significant predictors of LARFP use, including maternal education, household wealth, and geographic location( 12 ). While these methods have produced valuable results, they often can't depict the complex, nonlinear relationships between variables that influence family planning decisions. For instance, the interplay between education and access to health services, or geographic variations in service availability, may not be well represented by conventional statistical methods. Machine learning (ML) offers an exciting potential for interpreting complex datasets and uncovering hidden patterns that conventional methods would obfuscate ( 13 , 14 ). Unlike traditional statistical approaches, ML algorithms can model high-dimensional data and identify nonlinear relationships between variables ( 15 , 16 ). This makes ML well-placed to analyze the intricate determinants of LARFP use in Ethiopia. The newer application of ML has demonstrated its capability for enhancing predictive performance and informing actionable decisions in a variety of public health domains, including vaccine hesitancy, HIV testing, and maternal health ( 17 – 20 ). By leveraging the capabilities of ML, this study aims to transcend the limitations of traditional methods and obtain a more nuanced understanding of the determinants of LARFP use in Ethiopia. This study employs the 2021–2023 Performance Monitoring for Action (PMA) Ethiopia dataset, a nationally representative survey with in-depth data on reproductive health behaviors, including LARFP use( 21 ). The PMA dataset is particularly suitable for this research due to its granularity and because it has contextual variables, such as regional classification, household composition, and access to healthcare facilities. By applying ML models to this data, the study seeks to achieve three primary objectives: ( 1 ) identify the most significant determinants of LARFP use, ( 2 ) contrast the predictive power of ML models with traditional logistic regression, and ( 3 ) provide practical suggestions for targeted family planning interventions in Ethiopia. The outcome of this study can guide the formulation and execution of more effective family planning programs in Ethiopia. By comprehending the key determinants of LARFP use, policymakers can launch targeted initiatives that address the specific needs of subgroups in the population. For example, if the analysis reveals that the education of women is a significant intervention point to overcome financial constraints in access to LARFP, education measures could be included within family planning interventions. Similarly, if geographical inequalities are revealed to be an important bottleneck, investment can be allocated towards health infrastructure development in poorly served areas. Outside of Ethiopia, this study contributes to the nascent literature on the application of ML in global health, demonstrating its potential for enhancing the precision and efficacy of public health interventions. Overall, this study is an important contribution to the knowledge of the complex determinants of LARFP use in Ethiopia. By combining robust nationally representative data with state-of-the-art ML techniques, the study goes beyond the limitations of traditional statistical approaches and provides a more comprehensive decision-making tool in reproductive health. The results of this study can help bridge the gap between the availability and utilization of LARFP methods, which will ultimately translate into improved reproductive health outcomes and the success of Ethiopia's family planning agenda. Methods Data Source The analysis utilized the 2021–2023 Performance Monitoring for Action (PMA) Ethiopia dataset, which is a nationally representative survey established to monitor key reproductive health indicators. The dataset includes data on 9,763 women between the ages of 15 and 49 collected through a two-stage cluster sample design stratified by rural and urban residents. The survey provides comprehensive information on household population characteristics, reproductive health behaviors, and access to family planning services and thus constitutes an ideal source of data for LARFP usage analysis. Key Features of the Dataset Nationally Representative: The data are representative of the entire geographic area of Ethiopia, with the potential for extensive use of results. Broad Variables: It has variables like region, household member number, wealth quintile, availability of health care, and mother's level of education. Outcome Variable: is a feature generated (LARFP use) from the dataset, which is an intrauterine device (IUD) or implant utilization at survey. Variables By literature review available research, and analysis of datasets, 24 variables were found suitable and selected for this study. They are categorized under five thematic domains, each of which addresses specific issues of the study. The Geographic Context category has two variables: "region_cc," which captures Ethiopia's administrative regions to account for geographic variation in access to care, and "ur," a dummy variable capturing residence as urban or rural. The Socioeconomic & Household Conditions category has "lack_food_24h_4wk," a variable capturing household food insecurity as a function of whether the household had not eaten food for 24 hours or more in the previous four weeks. The third category, Healthcare Access & Utilization, contains variables related to healthcare encounters and family planning (FP) services. These are "health_check_6m_yn," a dummy variable indicating that the woman visited a health facility in the past six months; "fp_info_vaccine_visit_6m," which monitors whether FP information was given during vaccine visit in the same time frame; "fp_provider," details on the type of FP provider (e.g., clinic, community health worker); "talked_about_fp_6m," indicating whether FP was discussed with the provider or partner within the past six months; and "felt_encouraged," monitoring whether the woman was motivated to use FP during visits to health facilities. The fourth category, Family Planning Behavior & Perceptions, focuses on FP practices and decision-making. It includes "current_method", which identifies the contraceptive method currently in use, and "why_current_fp", a variable with sub-variables explaining the primary reason for choosing the current FP method. These sub-variables include "why_current_fp_duration" (duration of effectiveness), "why_current_fp_nofollowup" (minimal follow-up required), "why_current_fp_othersunavail" (other methods unavailable), "why_current_fp_recommendation" (provider recommendation), "why_current_fp_fewersidefx" (fewer side effects), "why_current_fp_ignoranthusband" (partner unaware), and "why_current_fp_other" (other reasons). Additionally, this category includes "fp_side_effects", indicating whether the individual experienced side effects from their FP method and "fp_told_switch", recording whether they were advised to switch FP methods due to side effects. The fifth category, Pregnancy & Reproductive Health, has "pregnancy_type" with which the pregnancy type is distinguished as intended or unintended, and "preg_now_react", which indicates the individual's response to the ongoing pregnancy (e.g., happy, worried). Finally, the sixth category, Reproductive Coercion & Partner Dynamics, addresses interpersonal determinants of reproductive choices. It includes "rc_forced_pregnancy_rw," which indicates if the respondent was coerced into becoming pregnant; "rc_partner_leave_rw," which measures partner abandonment as a factor considered in FP planning; and "last_sex_pressured," which reports coercion in last sex. Together, these measure a full analysis framework for multiple facets of FP determinants for family planning and reproductive health status. Data Balancing The class proportion of the output feature in the provided dataset leans slightly because the "others" class accounts for approximately 61.04% of data and the "long-term" class accounts for approximately 38.96% (Fig. 1 ). However, the trained machine learning model is functioning perfectly well with no impact from class imbalance, with remarkably high accuracy, precision, recall, and F1 measures for both classes. The functioning of the model to appropriately handle the class proportion is evident from its respective performance metrics. High precision means that the model is correctly classifying most of the instances in both classes. Precision and recall scores show that the model is not only accurate in prediction but also effective in identifying true positives with minimal false positives and false negatives. The F1 score, which combines precision and recall, also confirms the model's capability to address the class imbalance. With these strong performance metrics, no artificial balancing techniques such as oversampling or under sampling are required. The model is already working well on the class distribution and generating excellent results without any adjustments. This indicates that the model is ideally fitted to the task and can well predict outcomes for both the "others" and "long-term" classes. Data Preprocessing To prepare the dataset for analysis, the following preprocessing steps were undertaken. The dataset exhibited a low overall rate of missing values, with only 0.71% of records missing data in variables such as "fp_info_vaccine_visit_6m" and "fp_side_effects". Given the minimal and random distribution of missingness, these records were excluded to preserve data integrity and avoid introducing bias, as their removal was unlikely to substantially impact model performance. To prepare the dataset for machine learning analysis, various preprocessing steps were carried out on the provided variables. The categorical variables, i.e., "region_cc" and "ur" (urban/rural classification), were encoded with one-hot encoding. It converts the categorical variables into binary format and establishes separate columns for every category to avoid adding ordinal bias, which may introduce bias in machine learning models otherwise. For ordinal features such as "preg_now_react" (response to ongoing pregnancy) and "why_current_fp" (the primary reason for the use of ongoing family planning), min-max normalization was employed to standardize their values to 0 to 1. This preserves the relative ranking of ordinal values but also allows for equal scale for every feature. Besides, binary variables such as "health_check_6m_yn" (health facility visit in the past six months) and "fp_info_vaccine_visit_6m" (receipt of FP information during a vaccine visit) were not altered since they were already in an acceptable machine learning algorithm format. All these preprocessing operations, from one-hot encoding for categorical variables to normalization for ordinal variables, are carried out to ensure that the dataset is well formatted and scaled for robust machine learning analysis. Machine Learning Models To undertake this research, various machine learning models were utilized to compare performance in terms of predicting the target variable. Models were selected to cover a wide spectrum of algorithmic approaches in a comprehensive analysis of performance. Logistic Regression was used as one of the training models, a linear model for use in binary classification problems predicting the probability of a binary outcome with predictor variables. Random Forest, a class of ensemble learning that constructs many decision trees and outputs the class mode, was employed as well because of its stability and ability to prevent overfitting. XGBoost, an optimized gradient boosting library with high efficiency and flexibility, was also used to train models sequentially and correct mistakes made by previous iterations. Support Vector Machine (SVM), a strong supervised learning model, was used to identify the optimal hyperplane for the separation of classes. K-Nearest Neighbors (KNN), a simple yet effective model that classifies data points based on the majority class of their closest neighbors, was also used to provide a baseline comparison. These models were chosen to supply a complete and diverse evaluation of predictive ability across models. Model Training and Evaluation The models were evaluated using stratified 5-fold cross-validation such that class distribution (non-users of LARFP and users of LARFP) was maintained in each fold. It helps in the recovery of a better estimate of model performance, especially when there is a class imbalance. The dataset was split into an 80% training set and a 20% test set for each fold such that the models were trained and tested on distinct subsets of data. The training procedure began with the data preprocessing steps. Missing values in the data were handled by replacing them with "Unknown" to prevent any data loss. Categorical variables were encoded using LabelEncoder to convert them into a machine learning model usable form. The dataset was separated into features (X) and the target variable (y), where "LARFP use" was the target. For models that require standardization, such as Logistic Regression, SVM, KNN, and Naive Bayes, feature data was standardized using StandardScaler. This is so that all features contribute equally to the model training process by scaling their values. The outcomes were compiled into a data frame for easy comparison, and ROC curves for every model were plotted to observe how they fared in terms of true positive rate versus false positive rate. This integrated evaluation approach ensures a proper understanding of the advantages and disadvantages of each model in predicting the target variable. Performance Metrics To evaluate how well the machine learning algorithms are performing, several key metrics were employed to obtain a full representation of their ability to accurately predict the target variable, particularly for class imbalance conditions. Accuracy was used to calculate the number of correctly predicted instances divided by total instances to convey a broad sense of model performance. Accuracy was utilized to determine the proportion of true positive predictions out of all positive predictions the model produced, especially when false positives are expensive. Recall or sensitivity was utilized to determine the proportion of true positives identified out of all actual positives, which is important if false negatives are expensive. The F1 measure, which is the harmonic mean of recall and precision, was used to balance between the two measures, providing one measure that was equally sensitive to false positives and false negatives. The Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve was also computed to establish the ability of the model to distinguish between classes, with the higher the AUC value, the better. All these measures together form a complete assessment framework for measuring model performance in a balanced and advanced manner. Results Descriptive Statistics The analysis of the dataset revealed several significant results regarding the distribution and characteristics of the variables. The majority of the respondents (61.04%) were "others" in LARFP usage, and 38.96% were "long-term" users. Regionally, the highest percentage of respondents were from the Oromia region (30.24%), followed by SNNP (25.61%), Addis (22.52%), and Amhara (21.63%). The vast majority of the participants did not experience food scarcity in the last 24 hours of the last four weeks (98.12%), and most of the pregnancies were singleton (98.79%) compared to twin (1.21%). In medicine, 64.68% of the respondents never received a health check-up in the previous six months and 74.27% did not get family planning counseling on a vaccine visit during the same period. Emotional responses to pregnancy were complex, with 36.91% of the respondents reporting feeling "sort of unhappy" and 31.27% reporting feeling "very unhappy." Injections (46.58%) and implants (37.09%) were the most common methods of family planning. Justifications for choosing the current family planning method were fewer side effects (32.3%) and effectiveness period (15.5%). Family planning side effects were never felt by a majority of respondents (65.40%), and 56.64% were advised to switch methods. Government health centers (49.12%) and health posts (15.56%) were the principal providers of family planning. The majority of the respondents (82.52%) had not discussed family planning in the past six months, and 49.32% were motivated by their partners. Nearly all respondents (98.00%) had no coercive pregnancy, and 99.00% did not fear that their partner would leave them due to pregnancy. Furthermore, 92.48% had no coercion during the previous sexual intercourse. The sample was slightly more urban (55.41%) than rural (44.59%). These findings are high in information related to family planning and health behavior and demographic trends of the population being studied. The breakdown of long-term users of LARFP by region revealed there were great regional variations in adoption rates. Region SNNP recorded the highest proportion of long-term users at 30.59%, followed by Region Oromia at 28.05%, and Region Addis at 25.78% (Fig. 2 ). Region Amhara recorded the lowest with only 15.58% long-term users. These findings point to differential access to family planning services, knowledge, and cultural acceptability by region. The variations highlight the need for region-specific interventions in regions with low adoption rates to improve access to and knowledge of long-term family planning options. This regional analysis provides important lessons for policymakers to create and execute region-specific approaches to improve family planning outcomes. The analysis of the reasons for the choice of the current family planning method has some significant results. The most cited reason by the respondents is "fewer_side_fx" (fewer side effects), accounting for 32.3% of the responses. It can be seen that a significant percentage of users favor methods with fewer side effects. The second most common reason is "duration" (duration of effectiveness), reported by 15.5% of the respondents (Fig. 3 ). This shows that long-acting family planning methods requiring less intervention are liked by numerous users. The other notable reason is "no_follow_up" (no need for follow-up), reported by 15.0% of the respondents. This shows the need for methods that are not cumbersome and do not require frequent medical follow-ups. The other motivations include combinations such as "duration no_follow_up" (13.1%) and "fewer_side_fx no_follow_up" (4.5%) (Fig. 3 ). These combinations indicate the need for products that have both fewer side effects and no follow-up needs. Model Performance Table 1 Shows Trained Model Performance Model Accuracy Precision Recall F1 Score Logistic Regression 0.967033 0.972973 0.972973 0.972973 Random Forest 0.983516 0.990909 0.981982 0.986425 XGBoost 0.989011 1.000000 0.981982 0.990909 SVM 0.873626 0.900000 0.891892 0.895928 KNN 0.818681 0.848214 0.855856 0.852018 Naive Bayes 0.840659 0.910000 0.819820 0.862559 Decision Tree 0.994505 1.000000 0.990991 0.995475 The model comparison of the machine learning algorithms revealed that the Decision Tree model was the best among all, having the highest accuracy (99.45%), precision (100%), recall (99.10%), and F1 score (99.55%). This indicates its excellent ability to classify instances accurately and balance recall and precision. The XGBoost model also did well with an accuracy of 98.90%, precision of 100%, recall of 98.20%, and F1 score of 99.09%, closely followed by the Random Forest model with an accuracy of 98.35%, precision of 99.09%, recall of 98.20%, and F1 score of 98.64%. Logistic Regression yielded a good performance with 96.70% accuracy, 97.30% precision, 97.30% recall, and 97.30% F1 score. SVM, KNN, and Naive Bayes presented comparatively low performance, where SVM attained 87.36% accuracy, KNN 81.87%, and Naive Bayes 84.07% (Table 1 ). These results refer to the superiority of tree-based models (Decision Tree, XGBoost, and Random Forest) in handling the dataset, also revealing that the application of simpler models like Logistic Regression remains competitive. The results show that tree-based algorithms are extremely suitable for this classification task, with very good prediction performance and stability. The AUC-ROC curve is revealed by the AUC-ROC curve analysis to demonstrate that the Decision Tree model achieved an ideal AUC of 1.00, indicating flawless classification performance without misclassifications. This is subsequently followed by the Random Forest and XGBoost models, each with an AUC of 0.99, reflecting their strong ability to discriminate between classes. The Logistic Regression model was equally excellent with an AUC of 0.97, whereas the SVM model had an AUC of 0.96, which is a fair but slightly weaker performance. The Naive Bayes and KNN models also had relatively poorer AUC measures of 0.90 and 0.89, respectively, indicating less efficient classification performance (Fig. 4 ). These results are in line with the earlier accuracy, precision, recall, and F1 score outputs, affirming the superiority of tree-based models (Decision Tree, Random Forest, and XGBoost) for this classification task. The AUC-ROC curve results highlight the robustness and reliability of the models in handling the dataset, and thus they are appropriate choices for predictive tasks where high classification accuracy is desirable. The performance of the Decision Tree model indicates excellent classification power and balance between the two classes with an overall accuracy of 0.99. The classification report also indicates its strength with precision, recall, and F1-score of 0.99 or more for both classes. Specifically, for class 0, the model achieved a precision of 0.99, a recall of 1.00, and an F1-score of 0.99, whereas, for class 1, it achieved a perfect precision of 1.00, a recall of 0.99, and an F1-score of 1.00 (Fig. 5 ). This indicates that the model does exceedingly well in distinguishing instances belonging to both classes with minimal misclassifications. The balanced precision and recall show that the model is as good at handling class distribution as it is not biased towards the minority or majority class. This can also be evidenced by the perfect F1 scores, showing a perfect balance between precision and recall. The good performance per se does not need under-sampling or oversampling to offset class imbalance since the model handles distribution automatically. Also, the model's performance is already at its best, suggesting that there is no need to perform hyperparameter tuning. The Decision Tree's capability to generalize well and keep high performance in both classes ensures that it is a safe and effective option for this classification task. Feature Importance The Decision Tree model feature importance analysis reflects the relative contribution of every feature to the model's predictive performance. The most important feature is why_current_fp_duration, with an importance of 0.35, which indicates the significance of the current family planning method duration in determining the target variable. This is followed by fp_provider (0.25 importance score) and why_current_fp (0.20 importance score), which denote the important impact of the family planning service provider and the reasons for using the current method, respectively. Regional variation, as denoted by region_cc (0.15 importance score), is also critical in the model predictions, showing variations in family planning by geographical area. In addition, felt_encouraged (0.10 importance score) also indicates that encouragement by partners or clinicians is influencing decision-making. The preg_now_react, fp_side_effects, and health_check_6m_vn features are of medium importance, their saliency being significant but of lesser magnitude than the initial four features. Nonetheless, why_current_fp_ignoranthusband, pregnancy_type, and rc_partner_leave_rw features register small importance scores (close to 0.00), indicating minimal impact on the model's performance (Fig. 6 ). These findings underscore that the length of family planning procedures, providers, and reasons for method choice are the primary determinants of the model's choice, with others playing secondary or no roles. This examination helps ascertain the primary determinants of family planning decisions, which can be applied to inform focused interventions and policy advice. Discussion This study applied machine learning (ML) techniques to examine the use determinants of long-acting reversible family planning (LARFP) in Ethiopia, shedding fresh light on a major public health problem. The study overcame the limitations of traditional statistical methods by applying ML models to nationally representative data to uncover nuanced predictors and patterns and their potential application to inform precision interventions. What follows situates the findings within context, we reflect on the implications, limitations, and the future directions for the study. The Decision Tree model emerged as the most robust predictor of LARFP use, achieving near-perfect performance (accuracy: 99.45%, F1 score: 99.55%). This illustrates the ability of tree-based models to identify intricate, non-linear relationships in high-dimensional reproductive health data a task less readily accomplished by conventional logistic regression ( 22 , 23 ). The model's feature importance analysis revealed that the effectiveness duration of the chosen method (why_current_fp_duration) was the most important factor, followed by the potential FP provider type (fp_provider) and geographic region (region_cc). These findings are in line with existing literature emphasizing the role of method effectiveness and accessibility in FP utilization but extend them by quantifying the relative significance of these variables through ML( 24 , 25 ). For instance, regional disparities in the uptake of LARFP (for instance, 30.59% in SNNP, 15.58% in Amhara) confirm ethnographic results indicating cultural and infrastructural barriers in difficult-to-reach regions like Amhara. Notably, clinician and partner support and provider recommendation (“felt_encouraged”) were powerful influencers, consistent with socio-ecological models emphasizing interpersonal and institutional determinants that influence behavior. Conversely, other predictors such as reproductive coercion (“rc_partner_leave_rw”) and pregnancy type had no influence, opposite to qualitative findings pointing towards the influence of partners. This may be due to sensitive issues such as coercion being underreported in questionnaires or cultural details being narrowly framed in the dataset.xt, reflect on their implications, and address the study's limitations and the way forward. The results offer pragmatic recommendations to advance Ethiopia's family planning agenda using focused, multidimensional interventions. To begin, confronting dramatic regional differences in the use of LARFP (Fig. 2 ) requires regional-level interventions. In areas such as Amhara, where adoption is weakest (15.58%), investing in health facility expansion, training of providers, and community outreach education campaigns are essential to meet access gaps and overcome cultural challenges. Secondly, the rising prominence of fp_provider as a key determinant calls for provider-centric responses. Improving public health infrastructure government health facilities and health stations, the basic FP service institutions becomes essential ( 26 ). The training and inspiration of health personnel to proactively address side effect misconceptions and emphasize the benefits of LARFP interventions down the line will assist in bringing about client confidence and utilization ( 27 ). Third, demand-side strategies must concentrate on client-centric care as shown by the dominance of fewer side effects and absence of follow-up among the causes of method choice (Fig. 3 ). Expansion of access to less side effect-burdened newer LARFP technologies (e.g., hormonal implants) and integrating mobile health platforms to reduce follow-up burden can align services with user needs( 28 ). Finally, while partner coercion (“rc_partner_leave_rw”) had little influence, the signification of social support (“felt_encouraged”) highlights why partners and local leaders need to be engaged. Initiatives aiming to normalize FP discussions through programs of male engagement or community talks could establish favorable environments, drive away stigma, and enable women to make family-planning choices ( 28 , 29 ). All of these policies indicate the significance of a comprehensive strategy encompassing geographic, institutional, and socio-cultural aspects in a direction toward equitable access to family planning services. Advancing this research requires a multi-pronged agenda to deepen understanding and turn evidence into equitable health outcomes. Embedding mixed-methods approaches initially might strengthen the machine learning (ML) findings by placing them within qualitative evidence on cultural, religious, and gender-based barriers factors that are not well represented in the current dataset. Ethnographic studies, for instance, could elucidate why regions like Amhara lag in LARFP adoption, informing culturally sensitive interventions. Second, causal processes must be examined with longitudinal or quasi-experimental designs to disentangle temporal relationships between variables, such as how training providers influence LARFP use directly over time. Third, testing the generalizability of the model to other low-resource settings (e.g., sub-Saharan Africa or South Asia) could further refine transferable strategies by promoting scalability and considering contextual details such as health system fragmentation or socio-political dynamics. Finally, since ML is becoming more widely used in public health, rigorous ethical audits must explore algorithmic bias that could unintentionally disenfranchise vulnerable groups, such as rural communities or adolescents. Preventive measures such as inclusive data collection and participatory model development are imperative to ensure equitable technologies. All these pathways would harmonize technical innovation with socio-cultural needs, rendering precision public health technologies effective and ethical. Limitations of the Study The study's cross-sectional nature restricts causal inference because it prevents the assessment of temporal associations or controls for potential confounders across time. The data set also lacks detailed information on cultural beliefs and seasonal migration patterns, which limits control for how these dynamic and sociocultural variables can influence outcomes. Conclusion This study leveraged the power of machine learning to unpack the complex determinants of the utilization of long-acting reversible family planning (LARFP) in Ethiopia and offer a data-driven handbook to bridge the gap between availability and utilization. Drawing on nationally representative data from the 2021–2023 PMA Ethiopia survey, the study identified key predictors of LARFP uptake as the duration of method effectiveness, family planning provider source, and location, with wide variations observed across regions such as Amhara (15.58% uptake) and SNNP (30.59%). The superior performance of tree-based ML algorithms, particularly the Decision Tree algorithm (99.45% accuracy), demonstrated the strength of advanced analytics in identifying non-linear patterns and high-dimensional interactions that are often overlooked in traditional statistical analysis. These findings underscore the need for interventions that target geographic inequities, develop capacities in health systems, and align services with user preference e.g., prioritizing low-side-effect or minimal follow-up procedures. The study's results have international implications for precision public health. The integration of machine learning into reproductive health research allows policymakers to develop evidence-based, locally tailored strategies that respond to nuances at hand, ranging from scaling up provider capacity in far-flung regions to leveraging mobile health technologies for client-comprehensive care. Future research should prioritize responsible AI deployment, mixed-methods research to ground numerical results, and trans-regional collaboration to iterate on scalable models. As Ethiopia makes progress toward its FP2030 targets, this research illuminates the revolutionizing potential of evidence-based practice to bring family planning from a coveted policy objective to a tangible fact of life for millions of women. In the end, the path to universal access to LARFP is through reconciling technological advances with equitable, culturally tailored health systems a vision this study aims to unleash. Abbreviations AI: Artificial Intelligence AUC: Area Under the Curve FP: Family Planning FP2030: Family Planning 2030 (global partnership initiative) HIV: Human Immunodeficiency Virus IUD: Intrauterine Device KNN: K-Nearest Neighbors LARFP: Long-Acting Reversible Family Planning ML: Machine Learning PII: Personal Identifiable Information PMA: Performance Monitoring for Action ROC: Receiver Operating Characteristic SNNP: Southern Nations, Nationalities, and Peoples' Region (Ethiopia) SVM: Support Vector Machine XGBoost: Extreme Gradient Boosting Declarations Data Availability The datasets used and/or analyzed during the current study are available on the PMA website ( https://doi.org/10.34976/6vsc-6t49 ) (21). Ethics Approval and Consent to Participate The PMA Ethiopia survey dataset is a publicly accessible resource that complies with stringent ethical standards in line with the Declaration of Helsinki. Before data collection, informed consent was secured from all participating households. For illiterate respondents, witnessed verbal consent procedures were employed to ensure understanding and voluntary participation. To safeguard privacy, personally identifiable information (PII) was eliminated, and geographic identifiers were aggregated to the regional level to prevent the identification of individuals or specific communities. The study protocol was designed to ensure that no vulnerable populations, such as refugees or ethnic minorities, were disproportionately burdened or excluded. The PMA Ethiopia survey project officially authorized the use of the de-identified dataset through its legal registration and data access agreements. This open-access framework enhances transparency and reproducibility while ensuring the confidentiality of participants is maintained. Competing Interests The author declares no competing interests. Clinical Trial Number Not applicable Consent for Publication Not applicable. No identifying details, images, or personal information of participants are included in this manuscript. All data were anonymized before analysis, and no individual consent for publication was required. Funding No funding was received for this research. Acknowledgments The author thanks the PMA Ethiopia team for providing the dataset and the Ethiopian Public Health Institute for their support. References Kibret MA, Gebremedhin LT. Two decades of family planning in Ethiopia and the way forward to sustain hard-fought gains! Volume 19. Reproductive Health. BioMed Central Ltd; 2022. DeMaria LM, Smith KV, Berhane Y. Sexual and reproductive health in Ethiopia: gains and reflections over the past two decades. Volume 19. Reproductive Health. BioMed Central Ltd; 2022. World Family Planning. 2022 Meeting the changing needs for family planning: Contraceptive use by age and method. Weldekidan HA, Lemlem SB, Sinishaw Abebe W, Sori SA. Discontinuation rate of long-acting reversible contraceptives and associated factors among reproductive-age women in Butajira town, Central Ethiopia. Women’s Health. 2022;18. Arero WD, Teka WG, Hebo HJ, Woyo T, Amare B. Prevalence of long-acting reversible contraceptive methods utilization and associated factors among counseled mothers in immediate postpartum period at Jimma University medical center, Ethiopia. Contracept Reprod Med. 2022;7(1). Curtis KM, Peipert JF. Long-Acting Reversible Contraception. N Engl J Med. 2017;376(5):461–8. Damtew B, Yigezu M. Perspectives of care providers on obstacles to healthcare access for people with disabilities in Ethiopia: a qualitative investigation. BMC Health Serv Res. 2024;24(1):1290. Jisso M, Assefa NA, Alemayehu A, Gadisa A, Fikre R, Umer A, et al. Barriers to Family Planning Service Utilization in Ethiopia: A Qualitative Study. Ethiop J Health Sci. 2023;33(2):143–54. Bassoumah B, Adam AM, Adokiya MN. Challenges to the utilization of Community-based Health Planning and Services: the views of stakeholders in Yendi Municipality, Ghana. BMC Health Serv Res. 2021;21(1). Titiyos A, Mehretie Y, Alemayehu YK, Ejigu Y, Yitbarek K, Abraham Z et al. Family planning integration in Ethiopia’s primary health care system: a qualitative study on opportunities, challenges and best practices. Reprod Health. 2023;20(1). FP2030. ETHIOPIA GOVERNMENT COMMITMENT. Gudayu TW. Determinants of place birth: a multinomial logistic regression and spatial analysis of the Ethiopian mini demographic and health survey data, 2019. BMC Pregnancy Childbirth. 2022;22(1). Humphreys D, Kupresanin A, Boyer MD, Canik J, Chang CS, Cyr EC, et al. Advancing Fusion with Machine Learning Research Needs Workshop Report. J Fusion Energy. 2020;39(4):123–55. Peace P, Abayomi G, Bolla A. An In-Depth Analysis of Machine Learning Approaches: Methods, Applications, and Challenges. 2024. Aghaabbasi M, Chalermpong S. Machine learning techniques for evaluating the nonlinear link between built-environment characteristics and travel behaviors: A systematic review. Travel Behav Soc [Internet]. 2023;33:100640. Available from: https://www.sciencedirect.com/science/article/pii/S2214367X23000911 Ley C, Martin R, Pareek A, Groll A, Seil R, Tischer T. Machine learning and conventional statistics: making sense of the differences. Knee Surgery, Sports Traumatology, Arthroscopy. 2022;30:1–5. Ley C, Martin R, Pareek A, Groll A, Seil R, Tischer T. Machine learning and conventional statistics: making sense of the differences. Knee Surgery, Sports Traumatology, Arthroscopy. 2022;30:1–5. Jaiteh M, Phalane E, Shiferaw Y, Phaswana-Mafuya R. The Application of Machine Learning Algorithms to Predict HIV Testing in Repeated Adult Population–Based Surveys in South Africa: Protocol for a Multiwave Cross-Sectional Analysis. JMIR Res Protoc. 2025;14:e59916. Liang W, Chaudhary P, Rajuroy A, Hermange K, Yu Y. Machine Learning for Public Health Analytics. 2025. Liang W, Chaudhary P, Rajuroy A, Hermange K, Yu Y. Machine Learning for Public Health Analytics. 2025. Addis Ababa University School of Public Health, Ethiopia; Bill & Melinda Gates Institute for Population and Reproductive Health at the Johns Hopkins Bloomberg School of Public Health, USA. 2024, PMA Ethiopia Panel (2021–2023) Cohort 2: 6-month Follow-up Survey. https://doi.org/10.34976/6vsc-6t49 , Johns Hopkins Research Data Repository, V1. Kanyongo W, Ezugwu AE. Feature selection and importance of predictors of non-communicable diseases medication adherence from machine learning research perspectives. Inform Med Unlocked [Internet]. 2023;38:101232. Available from: https://www.sciencedirect.com/science/article/pii/S2352914823000746 Zhang Y, Chen Y, Su Q, Huang X, Li Q, Yang Y et al. The use of machine and deep learning to model the relationship between discomfort temperature and labor productivity loss among petrochemical workers. BMC Public Health. 2024;24(1). Titiyos A, Mehretie Y, Alemayehu YK, Ejigu Y, Yitbarek K, Abraham Z et al. Family planning integration in Ethiopia’s primary health care system: a qualitative study on opportunities, challenges and best practices. Reprod Health. 2023;20(1). Mulatu T, Sintayehu Y, Dessie Y, Dheresa M. Male involvement in family planning use and associated factors among currently married men in rural Eastern Ethiopia. SAGE Open Med. 2022;10. Ali D, Woldegiorgis AGY, Tilaye M, Yilma Y, Berhane HY, Tewahido D et al. Integrating private health facilities in government-led health systems: a case study of the public–private mix approach in Ethiopia. BMC Health Serv Res. 2022;22(1). Mwansisya T, Mbekenga C, Isangula K, Mwasha L, Mbelwa S, Lyimo M et al. The impact of training on self-reported performance in reproductive, maternal, and newborn health service delivery among healthcare workers in Tanzania: a baseline- and endline-survey. Reprod Health. 2022;19(1). Haddad LB, Townsend JW, Sitruk-Ware R. Contraceptive Technologies: Looking Ahead to New Approaches to Increase Options for Family Planning. Clin Obstet Gynecol. 2021;64(3):435–48. Silumbwe A, Nkole T, Munakampe MN, Cordero JP, Milford C, Zulu JM, et al. Facilitating community participation in family planning and contraceptive services provision and uptake: Community and health provider perspectives. Volume 17. Reproductive Health. BioMed Central Ltd; 2020. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 10 Jan, 2026 Read the published version in BMC Public Health → Version 1 posted Editorial decision: Revision requested 15 Oct, 2025 Reviews received at journal 14 Oct, 2025 Reviewers agreed at journal 26 Sep, 2025 Reviewers agreed at journal 11 Sep, 2025 Reviews received at journal 10 May, 2025 Reviewers agreed at journal 08 May, 2025 Reviewers agreed at journal 08 May, 2025 Reviewers agreed at journal 07 May, 2025 Reviewers invited by journal 29 Apr, 2025 Editor invited by journal 11 Apr, 2025 Editor assigned by journal 10 Apr, 2025 Submission checks completed at journal 10 Apr, 2025 First submitted to journal 10 Apr, 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-6417320","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":449677420,"identity":"b2764aa2-0a18-497c-b5f7-1eabf503fe67","order_by":0,"name":"Abraham Keffale Mengistu","email":"data:image/png;base64,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","orcid":"","institution":"Debre Markos University","correspondingAuthor":true,"prefix":"","firstName":"Abraham","middleName":"Keffale","lastName":"Mengistu","suffix":""}],"badges":[],"createdAt":"2025-04-10 07:08:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6417320/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6417320/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s12889-025-26158-7","type":"published","date":"2026-01-10T15:58:32+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":81703737,"identity":"70f1983d-f74d-4818-a4dd-d19c8b6c3cfa","added_by":"auto","created_at":"2025-04-30 13:12:00","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":37906,"visible":true,"origin":"","legend":"\u003cp\u003eClass Distribution of LARFP use\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-6417320/v1/e6b100243c12b143b328755b.png"},{"id":81703686,"identity":"13ced14f-43db-419e-8207-e80442f37ef7","added_by":"auto","created_at":"2025-04-30 13:11:55","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":42824,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of Long-term Users by Region\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-6417320/v1/95a5c5a6eebfeeaed17f39a4.png"},{"id":81703730,"identity":"bc45d707-fc99-45bd-a349-8e1948cf28ec","added_by":"auto","created_at":"2025-04-30 13:11:59","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":65200,"visible":true,"origin":"","legend":"\u003cp\u003eReasons for choosing the current FP method\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-6417320/v1/eeea03e9f0d768ee0a007532.png"},{"id":81703797,"identity":"76f6a03e-67d6-4d69-b641-931c6a6d3747","added_by":"auto","created_at":"2025-04-30 13:12:06","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":94204,"visible":true,"origin":"","legend":"\u003cp\u003eAUC-ROC curve plot of trained models\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-6417320/v1/b4caf25d71526ed3fba3008d.png"},{"id":81703760,"identity":"0dd9b0c8-3b7e-49aa-b999-5bdd5ae9b054","added_by":"auto","created_at":"2025-04-30 13:12:02","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":10431,"visible":true,"origin":"","legend":"\u003cp\u003eDecision tree model performance across different classes\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-6417320/v1/a20d807cedd8feba6273532a.png"},{"id":81703751,"identity":"446c0a17-8ee1-413e-86a5-93f8c6c9e650","added_by":"auto","created_at":"2025-04-30 13:12:01","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":100694,"visible":true,"origin":"","legend":"\u003cp\u003eFeature Importance for Decision Tree Model\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-6417320/v1/b2805e59f5ee3dc4e5bd153f.png"},{"id":100069318,"identity":"3abbe5fd-e92e-4497-8453-421cd0cb95a0","added_by":"auto","created_at":"2026-01-12 16:12:51","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":970250,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6417320/v1/67f4e027-cc49-432e-8696-362537e5db7c.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Exploring Machine Learning Insights into Long-Acting Reversible Family Planning Usage in Ethiopia: Analysis of the PMA (2021-2023) Dataset","fulltext":[{"header":"Introduction","content":"\u003cp\u003eEthiopia has made significant strides over the past decades in attaining better reproductive health outcomes, yet obstacles to universal access to family planning services remain (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e). Family planning is one of the pillars of reproductive health that enables individuals and couples to plan and have the number of children they desire and the space and time for their births(\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). Among the variety of family planning methods, long-acting reversible family planning (LARFP) methods such as intrauterine devices (IUDs) and implants are extremely effective in averting unintended pregnancy and maternal and child mortality(\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). These methods are particularly advantageous in low-resource contexts like Ethiopia, where health infrastructure and access to services are generally limited (\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Despite their effectiveness, the utilization of LARFPs in Ethiopia remains low, at only 22% among reproductive-age women (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e). The low utilization is attributed to a combination of socioeconomic, cultural, and access-to-healthcare barriers, which have hindered the widespread application of these life-saving interventions (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe Ethiopian government, in collaboration with international partners, has implemented various programs for the promotion of family planning, including the dissemination of LARFP methods(\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). However, the gap between their availability and use underscores the need to better understand their determinants of adoption (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). Traditional statistical approaches, such as logistic regression, have been widely used to identify significant predictors of LARFP use, including maternal education, household wealth, and geographic location(\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e). While these methods have produced valuable results, they often can't depict the complex, nonlinear relationships between variables that influence family planning decisions. For instance, the interplay between education and access to health services, or geographic variations in service availability, may not be well represented by conventional statistical methods.\u003c/p\u003e \u003cp\u003eMachine learning (ML) offers an exciting potential for interpreting complex datasets and uncovering hidden patterns that conventional methods would obfuscate (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e). Unlike traditional statistical approaches, ML algorithms can model high-dimensional data and identify nonlinear relationships between variables (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e). This makes ML well-placed to analyze the intricate determinants of LARFP use in Ethiopia. The newer application of ML has demonstrated its capability for enhancing predictive performance and informing actionable decisions in a variety of public health domains, including vaccine hesitancy, HIV testing, and maternal health (\u003cspan additionalcitationids=\"CR18 CR19\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e). By leveraging the capabilities of ML, this study aims to transcend the limitations of traditional methods and obtain a more nuanced understanding of the determinants of LARFP use in Ethiopia.\u003c/p\u003e \u003cp\u003eThis study employs the 2021\u0026ndash;2023 Performance Monitoring for Action (PMA) Ethiopia dataset, a nationally representative survey with in-depth data on reproductive health behaviors, including LARFP use(\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e). The PMA dataset is particularly suitable for this research due to its granularity and because it has contextual variables, such as regional classification, household composition, and access to healthcare facilities. By applying ML models to this data, the study seeks to achieve three primary objectives: (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) identify the most significant determinants of LARFP use, (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) contrast the predictive power of ML models with traditional logistic regression, and (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) provide practical suggestions for targeted family planning interventions in Ethiopia.\u003c/p\u003e \u003cp\u003eThe outcome of this study can guide the formulation and execution of more effective family planning programs in Ethiopia. By comprehending the key determinants of LARFP use, policymakers can launch targeted initiatives that address the specific needs of subgroups in the population. For example, if the analysis reveals that the education of women is a significant intervention point to overcome financial constraints in access to LARFP, education measures could be included within family planning interventions. Similarly, if geographical inequalities are revealed to be an important bottleneck, investment can be allocated towards health infrastructure development in poorly served areas. Outside of Ethiopia, this study contributes to the nascent literature on the application of ML in global health, demonstrating its potential for enhancing the precision and efficacy of public health interventions.\u003c/p\u003e \u003cp\u003eOverall, this study is an important contribution to the knowledge of the complex determinants of LARFP use in Ethiopia. By combining robust nationally representative data with state-of-the-art ML techniques, the study goes beyond the limitations of traditional statistical approaches and provides a more comprehensive decision-making tool in reproductive health. The results of this study can help bridge the gap between the availability and utilization of LARFP methods, which will ultimately translate into improved reproductive health outcomes and the success of Ethiopia's family planning agenda.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData Source\u003c/h2\u003e \u003cp\u003eThe analysis utilized the 2021\u0026ndash;2023 Performance Monitoring for Action (PMA) Ethiopia dataset, which is a nationally representative survey established to monitor key reproductive health indicators. The dataset includes data on 9,763 women between the ages of 15 and 49 collected through a two-stage cluster sample design stratified by rural and urban residents. The survey provides comprehensive information on household population characteristics, reproductive health behaviors, and access to family planning services and thus constitutes an ideal source of data for LARFP usage analysis.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eKey Features of the Dataset\u003c/h3\u003e\n\u003cp\u003eNationally Representative: The data are representative of the entire geographic area of Ethiopia, with the potential for extensive use of results.\u003c/p\u003e \u003cp\u003eBroad Variables: It has variables like region, household member number, wealth quintile, availability of health care, and mother's level of education.\u003c/p\u003e \u003cp\u003eOutcome Variable: is a feature generated (LARFP use) from the dataset, which is an intrauterine device (IUD) or implant utilization at survey.\u003c/p\u003e\n\u003ch3\u003eVariables\u003c/h3\u003e\n\u003cp\u003eBy literature review available research, and analysis of datasets, 24 variables were found suitable and selected for this study. They are categorized under five thematic domains, each of which addresses specific issues of the study. The Geographic Context category has two variables: \"region_cc,\" which captures Ethiopia's administrative regions to account for geographic variation in access to care, and \"ur,\" a dummy variable capturing residence as urban or rural. The Socioeconomic \u0026amp; Household Conditions category has \"lack_food_24h_4wk,\" a variable capturing household food insecurity as a function of whether the household had not eaten food for 24 hours or more in the previous four weeks.\u003c/p\u003e \u003cp\u003eThe third category, Healthcare Access \u0026amp; Utilization, contains variables related to healthcare encounters and family planning (FP) services. These are \"health_check_6m_yn,\" a dummy variable indicating that the woman visited a health facility in the past six months; \"fp_info_vaccine_visit_6m,\" which monitors whether FP information was given during vaccine visit in the same time frame; \"fp_provider,\" details on the type of FP provider (e.g., clinic, community health worker); \"talked_about_fp_6m,\" indicating whether FP was discussed with the provider or partner within the past six months; and \"felt_encouraged,\" monitoring whether the woman was motivated to use FP during visits to health facilities.\u003c/p\u003e \u003cp\u003eThe fourth category, Family Planning Behavior \u0026amp; Perceptions, focuses on FP practices and decision-making. It includes \"current_method\", which identifies the contraceptive method currently in use, and \"why_current_fp\", a variable with sub-variables explaining the primary reason for choosing the current FP method. These sub-variables include \"why_current_fp_duration\" (duration of effectiveness), \"why_current_fp_nofollowup\" (minimal follow-up required), \"why_current_fp_othersunavail\" (other methods unavailable), \"why_current_fp_recommendation\" (provider recommendation), \"why_current_fp_fewersidefx\" (fewer side effects), \"why_current_fp_ignoranthusband\" (partner unaware), and \"why_current_fp_other\" (other reasons). Additionally, this category includes \"fp_side_effects\", indicating whether the individual experienced side effects from their FP method and \"fp_told_switch\", recording whether they were advised to switch FP methods due to side effects.\u003c/p\u003e \u003cp\u003eThe fifth category, Pregnancy \u0026amp; Reproductive Health, has \"pregnancy_type\" with which the pregnancy type is distinguished as intended or unintended, and \"preg_now_react\", which indicates the individual's response to the ongoing pregnancy (e.g., happy, worried). Finally, the sixth category, Reproductive Coercion \u0026amp; Partner Dynamics, addresses interpersonal determinants of reproductive choices. It includes \"rc_forced_pregnancy_rw,\" which indicates if the respondent was coerced into becoming pregnant; \"rc_partner_leave_rw,\" which measures partner abandonment as a factor considered in FP planning; and \"last_sex_pressured,\" which reports coercion in last sex. Together, these measure a full analysis framework for multiple facets of FP determinants for family planning and reproductive health status.\u003c/p\u003e\n\u003ch3\u003eData Balancing\u003c/h3\u003e\n\u003cp\u003eThe class proportion of the output feature in the provided dataset leans slightly because the \"others\" class accounts for approximately 61.04% of data and the \"long-term\" class accounts for approximately 38.96% (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). However, the trained machine learning model is functioning perfectly well with no impact from class imbalance, with remarkably high accuracy, precision, recall, and F1 measures for both classes.\u003c/p\u003e \u003cp\u003eThe functioning of the model to appropriately handle the class proportion is evident from its respective performance metrics. High precision means that the model is correctly classifying most of the instances in both classes. Precision and recall scores show that the model is not only accurate in prediction but also effective in identifying true positives with minimal false positives and false negatives. The F1 score, which combines precision and recall, also confirms the model's capability to address the class imbalance.\u003c/p\u003e \u003cp\u003eWith these strong performance metrics, no artificial balancing techniques such as oversampling or under sampling are required. The model is already working well on the class distribution and generating excellent results without any adjustments. This indicates that the model is ideally fitted to the task and can well predict outcomes for both the \"others\" and \"long-term\" classes.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eData Preprocessing\u003c/h3\u003e\n\u003cp\u003eTo prepare the dataset for analysis, the following preprocessing steps were undertaken. The dataset exhibited a low overall rate of missing values, with only 0.71% of records missing data in variables such as \"fp_info_vaccine_visit_6m\" and \"fp_side_effects\". Given the minimal and random distribution of missingness, these records were excluded to preserve data integrity and avoid introducing bias, as their removal was unlikely to substantially impact model performance.\u003c/p\u003e \u003cp\u003eTo prepare the dataset for machine learning analysis, various preprocessing steps were carried out on the provided variables. The categorical variables, i.e., \"region_cc\" and \"ur\" (urban/rural classification), were encoded with one-hot encoding. It converts the categorical variables into binary format and establishes separate columns for every category to avoid adding ordinal bias, which may introduce bias in machine learning models otherwise.\u003c/p\u003e \u003cp\u003eFor ordinal features such as \"preg_now_react\" (response to ongoing pregnancy) and \"why_current_fp\" (the primary reason for the use of ongoing family planning), min-max normalization was employed to standardize their values to 0 to 1. This preserves the relative ranking of ordinal values but also allows for equal scale for every feature. Besides, binary variables such as \"health_check_6m_yn\" (health facility visit in the past six months) and \"fp_info_vaccine_visit_6m\" (receipt of FP information during a vaccine visit) were not altered since they were already in an acceptable machine learning algorithm format.\u003c/p\u003e \u003cp\u003eAll these preprocessing operations, from one-hot encoding for categorical variables to normalization for ordinal variables, are carried out to ensure that the dataset is well formatted and scaled for robust machine learning analysis.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eMachine Learning Models\u003c/h2\u003e \u003cp\u003eTo undertake this research, various machine learning models were utilized to compare performance in terms of predicting the target variable. Models were selected to cover a wide spectrum of algorithmic approaches in a comprehensive analysis of performance. Logistic Regression was used as one of the training models, a linear model for use in binary classification problems predicting the probability of a binary outcome with predictor variables. Random Forest, a class of ensemble learning that constructs many decision trees and outputs the class mode, was employed as well because of its stability and ability to prevent overfitting. XGBoost, an optimized gradient boosting library with high efficiency and flexibility, was also used to train models sequentially and correct mistakes made by previous iterations. Support Vector Machine (SVM), a strong supervised learning model, was used to identify the optimal hyperplane for the separation of classes. K-Nearest Neighbors (KNN), a simple yet effective model that classifies data points based on the majority class of their closest neighbors, was also used to provide a baseline comparison. These models were chosen to supply a complete and diverse evaluation of predictive ability across models.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eModel Training and Evaluation\u003c/h3\u003e\n\u003cp\u003eThe models were evaluated using stratified 5-fold cross-validation such that class distribution (non-users of LARFP and users of LARFP) was maintained in each fold. It helps in the recovery of a better estimate of model performance, especially when there is a class imbalance. The dataset was split into an 80% training set and a 20% test set for each fold such that the models were trained and tested on distinct subsets of data.\u003c/p\u003e \u003cp\u003eThe training procedure began with the data preprocessing steps. Missing values in the data were handled by replacing them with \"Unknown\" to prevent any data loss. Categorical variables were encoded using LabelEncoder to convert them into a machine learning model usable form. The dataset was separated into features (X) and the target variable (y), where \"LARFP use\" was the target.\u003c/p\u003e \u003cp\u003eFor models that require standardization, such as Logistic Regression, SVM, KNN, and Naive Bayes, feature data was standardized using StandardScaler. This is so that all features contribute equally to the model training process by scaling their values.\u003c/p\u003e \u003cp\u003eThe outcomes were compiled into a data frame for easy comparison, and ROC curves for every model were plotted to observe how they fared in terms of true positive rate versus false positive rate. This integrated evaluation approach ensures a proper understanding of the advantages and disadvantages of each model in predicting the target variable.\u003c/p\u003e\n\u003ch3\u003ePerformance Metrics\u003c/h3\u003e\n\u003cp\u003eTo evaluate how well the machine learning algorithms are performing, several key metrics were employed to obtain a full representation of their ability to accurately predict the target variable, particularly for class imbalance conditions. Accuracy was used to calculate the number of correctly predicted instances divided by total instances to convey a broad sense of model performance. Accuracy was utilized to determine the proportion of true positive predictions out of all positive predictions the model produced, especially when false positives are expensive. Recall or sensitivity was utilized to determine the proportion of true positives identified out of all actual positives, which is important if false negatives are expensive. The F1 measure, which is the harmonic mean of recall and precision, was used to balance between the two measures, providing one measure that was equally sensitive to false positives and false negatives. The Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve was also computed to establish the ability of the model to distinguish between classes, with the higher the AUC value, the better. All these measures together form a complete assessment framework for measuring model performance in a balanced and advanced manner.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eDescriptive Statistics\u003c/h2\u003e \u003cp\u003eThe analysis of the dataset revealed several significant results regarding the distribution and characteristics of the variables. The majority of the respondents (61.04%) were \"others\" in LARFP usage, and 38.96% were \"long-term\" users. Regionally, the highest percentage of respondents were from the Oromia region (30.24%), followed by SNNP (25.61%), Addis (22.52%), and Amhara (21.63%). The vast majority of the participants did not experience food scarcity in the last 24 hours of the last four weeks (98.12%), and most of the pregnancies were singleton (98.79%) compared to twin (1.21%).\u003c/p\u003e \u003cp\u003eIn medicine, 64.68% of the respondents never received a health check-up in the previous six months and 74.27% did not get family planning counseling on a vaccine visit during the same period. Emotional responses to pregnancy were complex, with 36.91% of the respondents reporting feeling \"sort of unhappy\" and 31.27% reporting feeling \"very unhappy.\" Injections (46.58%) and implants (37.09%) were the most common methods of family planning.\u003c/p\u003e \u003cp\u003eJustifications for choosing the current family planning method were fewer side effects (32.3%) and effectiveness period (15.5%). Family planning side effects were never felt by a majority of respondents (65.40%), and 56.64% were advised to switch methods. Government health centers (49.12%) and health posts (15.56%) were the principal providers of family planning.\u003c/p\u003e \u003cp\u003eThe majority of the respondents (82.52%) had not discussed family planning in the past six months, and 49.32% were motivated by their partners. Nearly all respondents (98.00%) had no coercive pregnancy, and 99.00% did not fear that their partner would leave them due to pregnancy. Furthermore, 92.48% had no coercion during the previous sexual intercourse. The sample was slightly more urban (55.41%) than rural (44.59%). These findings are high in information related to family planning and health behavior and demographic trends of the population being studied.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe breakdown of long-term users of LARFP by region revealed there were great regional variations in adoption rates. Region SNNP recorded the highest proportion of long-term users at 30.59%, followed by Region Oromia at 28.05%, and Region Addis at 25.78% (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Region Amhara recorded the lowest with only 15.58% long-term users. These findings point to differential access to family planning services, knowledge, and cultural acceptability by region. The variations highlight the need for region-specific interventions in regions with low adoption rates to improve access to and knowledge of long-term family planning options. This regional analysis provides important lessons for policymakers to create and execute region-specific approaches to improve family planning outcomes.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe analysis of the reasons for the choice of the current family planning method has some significant results. The most cited reason by the respondents is \"fewer_side_fx\" (fewer side effects), accounting for 32.3% of the responses. It can be seen that a significant percentage of users favor methods with fewer side effects. The second most common reason is \"duration\" (duration of effectiveness), reported by 15.5% of the respondents (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). This shows that long-acting family planning methods requiring less intervention are liked by numerous users.\u003c/p\u003e \u003cp\u003eThe other notable reason is \"no_follow_up\" (no need for follow-up), reported by 15.0% of the respondents. This shows the need for methods that are not cumbersome and do not require frequent medical follow-ups. The other motivations include combinations such as \"duration no_follow_up\" (13.1%) and \"fewer_side_fx no_follow_up\" (4.5%) (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). These combinations indicate the need for products that have both fewer side effects and no follow-up needs.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eModel Performance\u003c/h2\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eShows Trained Model Performance\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1 Score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLogistic Regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.967033\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.972973\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.972973\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.972973\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.983516\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.990909\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.981982\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.986425\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.989011\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.000000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.981982\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.990909\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.873626\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.900000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.891892\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.895928\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKNN\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.818681\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.848214\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.855856\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.852018\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNaive Bayes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.840659\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.910000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.819820\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.862559\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDecision Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.994505\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.000000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.990991\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.995475\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe model comparison of the machine learning algorithms revealed that the Decision Tree model was the best among all, having the highest accuracy (99.45%), precision (100%), recall (99.10%), and F1 score (99.55%). This indicates its excellent ability to classify instances accurately and balance recall and precision. The XGBoost model also did well with an accuracy of 98.90%, precision of 100%, recall of 98.20%, and F1 score of 99.09%, closely followed by the Random Forest model with an accuracy of 98.35%, precision of 99.09%, recall of 98.20%, and F1 score of 98.64%. Logistic Regression yielded a good performance with 96.70% accuracy, 97.30% precision, 97.30% recall, and 97.30% F1 score. SVM, KNN, and Naive Bayes presented comparatively low performance, where SVM attained 87.36% accuracy, KNN 81.87%, and Naive Bayes 84.07% (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). These results refer to the superiority of tree-based models (Decision Tree, XGBoost, and Random Forest) in handling the dataset, also revealing that the application of simpler models like Logistic Regression remains competitive. The results show that tree-based algorithms are extremely suitable for this classification task, with very good prediction performance and stability.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe AUC-ROC curve is revealed by the AUC-ROC curve analysis to demonstrate that the Decision Tree model achieved an ideal AUC of 1.00, indicating flawless classification performance without misclassifications. This is subsequently followed by the Random Forest and XGBoost models, each with an AUC of 0.99, reflecting their strong ability to discriminate between classes. The Logistic Regression model was equally excellent with an AUC of 0.97, whereas the SVM model had an AUC of 0.96, which is a fair but slightly weaker performance. The Naive Bayes and KNN models also had relatively poorer AUC measures of 0.90 and 0.89, respectively, indicating less efficient classification performance (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThese results are in line with the earlier accuracy, precision, recall, and F1 score outputs, affirming the superiority of tree-based models (Decision Tree, Random Forest, and XGBoost) for this classification task. The AUC-ROC curve results highlight the robustness and reliability of the models in handling the dataset, and thus they are appropriate choices for predictive tasks where high classification accuracy is desirable.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe performance of the Decision Tree model indicates excellent classification power and balance between the two classes with an overall accuracy of 0.99. The classification report also indicates its strength with precision, recall, and F1-score of 0.99 or more for both classes. Specifically, for class 0, the model achieved a precision of 0.99, a recall of 1.00, and an F1-score of 0.99, whereas, for class 1, it achieved a perfect precision of 1.00, a recall of 0.99, and an F1-score of 1.00 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). This indicates that the model does exceedingly well in distinguishing instances belonging to both classes with minimal misclassifications.\u003c/p\u003e \u003cp\u003eThe balanced precision and recall show that the model is as good at handling class distribution as it is not biased towards the minority or majority class. This can also be evidenced by the perfect F1 scores, showing a perfect balance between precision and recall. The good performance per se does not need under-sampling or oversampling to offset class imbalance since the model handles distribution automatically. Also, the model's performance is already at its best, suggesting that there is no need to perform hyperparameter tuning. The Decision Tree's capability to generalize well and keep high performance in both classes ensures that it is a safe and effective option for this classification task.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eFeature Importance\u003c/h2\u003e \u003cp\u003eThe Decision Tree model feature importance analysis reflects the relative contribution of every feature to the model's predictive performance. The most important feature is why_current_fp_duration, with an importance of 0.35, which indicates the significance of the current family planning method duration in determining the target variable. This is followed by fp_provider (0.25 importance score) and why_current_fp (0.20 importance score), which denote the important impact of the family planning service provider and the reasons for using the current method, respectively. Regional variation, as denoted by region_cc (0.15 importance score), is also critical in the model predictions, showing variations in family planning by geographical area. In addition, felt_encouraged (0.10 importance score) also indicates that encouragement by partners or clinicians is influencing decision-making. The preg_now_react, fp_side_effects, and health_check_6m_vn features are of medium importance, their saliency being significant but of lesser magnitude than the initial four features. Nonetheless, why_current_fp_ignoranthusband, pregnancy_type, and rc_partner_leave_rw features register small importance scores (close to 0.00), indicating minimal impact on the model's performance (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). These findings underscore that the length of family planning procedures, providers, and reasons for method choice are the primary determinants of the model's choice, with others playing secondary or no roles. This examination helps ascertain the primary determinants of family planning decisions, which can be applied to inform focused interventions and policy advice.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study applied machine learning (ML) techniques to examine the use determinants of long-acting reversible family planning (LARFP) in Ethiopia, shedding fresh light on a major public health problem. The study overcame the limitations of traditional statistical methods by applying ML models to nationally representative data to uncover nuanced predictors and patterns and their potential application to inform precision interventions. What follows situates the findings within context, we reflect on the implications, limitations, and the future directions for the study.\u003c/p\u003e \u003cp\u003eThe Decision Tree model emerged as the most robust predictor of LARFP use, achieving near-perfect performance (accuracy: 99.45%, F1 score: 99.55%). This illustrates the ability of tree-based models to identify intricate, non-linear relationships in high-dimensional reproductive health data a task less readily accomplished by conventional logistic regression (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e). The model's feature importance analysis revealed that the effectiveness duration of the chosen method (why_current_fp_duration) was the most important factor, followed by the potential FP provider type (fp_provider) and geographic region (region_cc). These findings are in line with existing literature emphasizing the role of method effectiveness and accessibility in FP utilization but extend them by quantifying the relative significance of these variables through ML(\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). For instance, regional disparities in the uptake of LARFP (for instance, 30.59% in SNNP, 15.58% in Amhara) confirm ethnographic results indicating cultural and infrastructural barriers in difficult-to-reach regions like Amhara.\u003c/p\u003e \u003cp\u003eNotably, clinician and partner support and provider recommendation (\u0026ldquo;felt_encouraged\u0026rdquo;) were powerful influencers, consistent with socio-ecological models emphasizing interpersonal and institutional determinants that influence behavior. Conversely, other predictors such as reproductive coercion (\u0026ldquo;rc_partner_leave_rw\u0026rdquo;) and pregnancy type had no influence, opposite to qualitative findings pointing towards the influence of partners. This may be due to sensitive issues such as coercion being underreported in questionnaires or cultural details being narrowly framed in the dataset.xt, reflect on their implications, and address the study's limitations and the way forward.\u003c/p\u003e \u003cp\u003eThe results offer pragmatic recommendations to advance Ethiopia's family planning agenda using focused, multidimensional interventions. To begin, confronting dramatic regional differences in the use of LARFP (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e) requires regional-level interventions. In areas such as Amhara, where adoption is weakest (15.58%), investing in health facility expansion, training of providers, and community outreach education campaigns are essential to meet access gaps and overcome cultural challenges. Secondly, the rising prominence of fp_provider as a key determinant calls for provider-centric responses. Improving public health infrastructure government health facilities and health stations, the basic FP service institutions becomes essential (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e). The training and inspiration of health personnel to proactively address side effect misconceptions and emphasize the benefits of LARFP interventions down the line will assist in bringing about client confidence and utilization (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e). Third, demand-side strategies must concentrate on client-centric care as shown by the dominance of fewer side effects and absence of follow-up among the causes of method choice (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Expansion of access to less side effect-burdened newer LARFP technologies (e.g., hormonal implants) and integrating mobile health platforms to reduce follow-up burden can align services with user needs(\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e). Finally, while partner coercion (\u0026ldquo;rc_partner_leave_rw\u0026rdquo;) had little influence, the signification of social support (\u0026ldquo;felt_encouraged\u0026rdquo;) highlights why partners and local leaders need to be engaged. Initiatives aiming to normalize FP discussions through programs of male engagement or community talks could establish favorable environments, drive away stigma, and enable women to make family-planning choices (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e). All of these policies indicate the significance of a comprehensive strategy encompassing geographic, institutional, and socio-cultural aspects in a direction toward equitable access to family planning services.\u003c/p\u003e \u003cp\u003eAdvancing this research requires a multi-pronged agenda to deepen understanding and turn evidence into equitable health outcomes. Embedding mixed-methods approaches initially might strengthen the machine learning (ML) findings by placing them within qualitative evidence on cultural, religious, and gender-based barriers factors that are not well represented in the current dataset. Ethnographic studies, for instance, could elucidate why regions like Amhara lag in LARFP adoption, informing culturally sensitive interventions. Second, causal processes must be examined with longitudinal or quasi-experimental designs to disentangle temporal relationships between variables, such as how training providers influence LARFP use directly over time. Third, testing the generalizability of the model to other low-resource settings (e.g., sub-Saharan Africa or South Asia) could further refine transferable strategies by promoting scalability and considering contextual details such as health system fragmentation or socio-political dynamics. Finally, since ML is becoming more widely used in public health, rigorous ethical audits must explore algorithmic bias that could unintentionally disenfranchise vulnerable groups, such as rural communities or adolescents. Preventive measures such as inclusive data collection and participatory model development are imperative to ensure equitable technologies. All these pathways would harmonize technical innovation with socio-cultural needs, rendering precision public health technologies effective and ethical.\u003c/p\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eLimitations of the Study\u003c/h2\u003e \u003cp\u003eThe study's cross-sectional nature restricts causal inference because it prevents the assessment of temporal associations or controls for potential confounders across time. The data set also lacks detailed information on cultural beliefs and seasonal migration patterns, which limits control for how these dynamic and sociocultural variables can influence outcomes.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study leveraged the power of machine learning to unpack the complex determinants of the utilization of long-acting reversible family planning (LARFP) in Ethiopia and offer a data-driven handbook to bridge the gap between availability and utilization. Drawing on nationally representative data from the 2021\u0026ndash;2023 PMA Ethiopia survey, the study identified key predictors of LARFP uptake as the duration of method effectiveness, family planning provider source, and location, with wide variations observed across regions such as Amhara (15.58% uptake) and SNNP (30.59%). The superior performance of tree-based ML algorithms, particularly the Decision Tree algorithm (99.45% accuracy), demonstrated the strength of advanced analytics in identifying non-linear patterns and high-dimensional interactions that are often overlooked in traditional statistical analysis. These findings underscore the need for interventions that target geographic inequities, develop capacities in health systems, and align services with user preference e.g., prioritizing low-side-effect or minimal follow-up procedures.\u003c/p\u003e \u003cp\u003eThe study's results have international implications for precision public health. The integration of machine learning into reproductive health research allows policymakers to develop evidence-based, locally tailored strategies that respond to nuances at hand, ranging from scaling up provider capacity in far-flung regions to leveraging mobile health technologies for client-comprehensive care. Future research should prioritize responsible AI deployment, mixed-methods research to ground numerical results, and trans-regional collaboration to iterate on scalable models. As Ethiopia makes progress toward its FP2030 targets, this research illuminates the revolutionizing potential of evidence-based practice to bring family planning from a coveted policy objective to a tangible fact of life for millions of women. In the end, the path to universal access to LARFP is through reconciling technological advances with equitable, culturally tailored health systems a vision this study aims to unleash.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eAI: Artificial Intelligence\u003c/p\u003e\n\u003cp\u003eAUC: Area Under the Curve\u003c/p\u003e\n\u003cp\u003eFP: Family Planning\u003c/p\u003e\n\u003cp\u003eFP2030: Family Planning 2030 (global partnership initiative)\u003c/p\u003e\n\u003cp\u003eHIV: Human Immunodeficiency Virus\u003c/p\u003e\n\u003cp\u003eIUD: Intrauterine Device\u003c/p\u003e\n\u003cp\u003eKNN: K-Nearest Neighbors\u003c/p\u003e\n\u003cp\u003eLARFP: Long-Acting Reversible Family Planning\u003c/p\u003e\n\u003cp\u003eML: Machine Learning\u003c/p\u003e\n\u003cp\u003ePII: Personal Identifiable Information\u003c/p\u003e\n\u003cp\u003ePMA: Performance Monitoring for Action\u003c/p\u003e\n\u003cp\u003eROC: Receiver Operating Characteristic\u003c/p\u003e\n\u003cp\u003eSNNP: Southern Nations, Nationalities, and Peoples' Region (Ethiopia)\u003c/p\u003e\n\u003cp\u003eSVM: Support Vector Machine\u003c/p\u003e\n\u003cp\u003eXGBoost: Extreme Gradient Boosting\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe datasets used and/or analyzed during the current study are available on the PMA website\u0026nbsp;(\u0026nbsp;\u003cu\u003ehttps://doi.org/10.34976/6vsc-6t49\u003c/u\u003e) (21).\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eEthics Approval and Consent to Participate\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe PMA Ethiopia survey dataset is a publicly accessible resource that complies with stringent ethical standards in line with the Declaration of Helsinki. Before data collection, informed consent was secured from all participating households. For illiterate respondents, witnessed verbal consent procedures were employed to ensure understanding and voluntary participation. To safeguard privacy, personally identifiable information (PII) was eliminated, and geographic identifiers were aggregated to the regional level to prevent the identification of individuals or specific communities. The study protocol was designed to ensure that no vulnerable populations, such as refugees or ethnic minorities, were disproportionately burdened or excluded. The PMA Ethiopia survey project officially authorized the use of the de-identified dataset through its legal registration and data access agreements. This open-access framework enhances transparency and reproducibility while ensuring the confidentiality of participants is maintained.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author declares no competing interests.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eClinical Trial Number\u0026nbsp;\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eConsent for Publication\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eNot applicable. No identifying details, images, or personal information of participants are included in this manuscript. All data were anonymized before analysis, and no individual consent for publication was required.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eNo funding was received for this research.\u003c/p\u003e\n\u003ch2\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/h2\u003e\n\u003cp\u003eThe author thanks the PMA Ethiopia team for providing the dataset and the Ethiopian Public Health Institute for their support.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eKibret MA, Gebremedhin LT. Two decades of family planning in Ethiopia and the way forward to sustain hard-fought gains! Volume 19. Reproductive Health. BioMed Central Ltd; 2022.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeMaria LM, Smith KV, Berhane Y. Sexual and reproductive health in Ethiopia: gains and reflections over the past two decades. Volume 19. Reproductive Health. BioMed Central Ltd; 2022.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Family Planning. 2022 Meeting the changing needs for family planning: Contraceptive use by age and method.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeldekidan HA, Lemlem SB, Sinishaw Abebe W, Sori SA. Discontinuation rate of long-acting reversible contraceptives and associated factors among reproductive-age women in Butajira town, Central Ethiopia. Women\u0026rsquo;s Health. 2022;18.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eArero WD, Teka WG, Hebo HJ, Woyo T, Amare B. Prevalence of long-acting reversible contraceptive methods utilization and associated factors among counseled mothers in immediate postpartum period at Jimma University medical center, Ethiopia. Contracept Reprod Med. 2022;7(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCurtis KM, Peipert JF. Long-Acting Reversible Contraception. N Engl J Med. 2017;376(5):461\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDamtew B, Yigezu M. Perspectives of care providers on obstacles to healthcare access for people with disabilities in Ethiopia: a qualitative investigation. BMC Health Serv Res. 2024;24(1):1290.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJisso M, Assefa NA, Alemayehu A, Gadisa A, Fikre R, Umer A, et al. Barriers to Family Planning Service Utilization in Ethiopia: A Qualitative Study. Ethiop J Health Sci. 2023;33(2):143\u0026ndash;54.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBassoumah B, Adam AM, Adokiya MN. Challenges to the utilization of Community-based Health Planning and Services: the views of stakeholders in Yendi Municipality, Ghana. BMC Health Serv Res. 2021;21(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTitiyos A, Mehretie Y, Alemayehu YK, Ejigu Y, Yitbarek K, Abraham Z et al. Family planning integration in Ethiopia\u0026rsquo;s primary health care system: a qualitative study on opportunities, challenges and best practices. Reprod Health. 2023;20(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFP2030. ETHIOPIA GOVERNMENT COMMITMENT.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGudayu TW. Determinants of place birth: a multinomial logistic regression and spatial analysis of the Ethiopian mini demographic and health survey data, 2019. BMC Pregnancy Childbirth. 2022;22(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHumphreys D, Kupresanin A, Boyer MD, Canik J, Chang CS, Cyr EC, et al. Advancing Fusion with Machine Learning Research Needs Workshop Report. J Fusion Energy. 2020;39(4):123\u0026ndash;55.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeace P, Abayomi G, Bolla A. An In-Depth Analysis of Machine Learning Approaches: Methods, Applications, and Challenges. 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAghaabbasi M, Chalermpong S. Machine learning techniques for evaluating the nonlinear link between built-environment characteristics and travel behaviors: A systematic review. Travel Behav Soc [Internet]. 2023;33:100640. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.sciencedirect.com/science/article/pii/S2214367X23000911\u003c/span\u003e\u003cspan address=\"https://www.sciencedirect.com/science/article/pii/S2214367X23000911\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLey C, Martin R, Pareek A, Groll A, Seil R, Tischer T. Machine learning and conventional statistics: making sense of the differences. Knee Surgery, Sports Traumatology, Arthroscopy. 2022;30:1\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLey C, Martin R, Pareek A, Groll A, Seil R, Tischer T. Machine learning and conventional statistics: making sense of the differences. Knee Surgery, Sports Traumatology, Arthroscopy. 2022;30:1\u0026ndash;5.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJaiteh M, Phalane E, Shiferaw Y, Phaswana-Mafuya R. The Application of Machine Learning Algorithms to Predict HIV Testing in Repeated Adult Population\u0026ndash;Based Surveys in South Africa: Protocol for a Multiwave Cross-Sectional Analysis. JMIR Res Protoc. 2025;14:e59916.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiang W, Chaudhary P, Rajuroy A, Hermange K, Yu Y. Machine Learning for Public Health Analytics. 2025.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiang W, Chaudhary P, Rajuroy A, Hermange K, Yu Y. Machine Learning for Public Health Analytics. 2025.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAddis Ababa University School of Public Health, Ethiopia; Bill \u0026amp; Melinda Gates Institute for Population and Reproductive Health at the Johns Hopkins Bloomberg School of Public Health, USA. 2024, PMA Ethiopia Panel (2021\u0026ndash;2023) Cohort 2: 6-month Follow-up Survey. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.34976/6vsc-6t49\u003c/span\u003e\u003cspan address=\"10.34976/6vsc-6t49\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e, Johns Hopkins Research Data Repository, V1.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKanyongo W, Ezugwu AE. Feature selection and importance of predictors of non-communicable diseases medication adherence from machine learning research perspectives. Inform Med Unlocked [Internet]. 2023;38:101232. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.sciencedirect.com/science/article/pii/S2352914823000746\u003c/span\u003e\u003cspan address=\"https://www.sciencedirect.com/science/article/pii/S2352914823000746\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Y, Chen Y, Su Q, Huang X, Li Q, Yang Y et al. The use of machine and deep learning to model the relationship between discomfort temperature and labor productivity loss among petrochemical workers. BMC Public Health. 2024;24(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTitiyos A, Mehretie Y, Alemayehu YK, Ejigu Y, Yitbarek K, Abraham Z et al. Family planning integration in Ethiopia\u0026rsquo;s primary health care system: a qualitative study on opportunities, challenges and best practices. Reprod Health. 2023;20(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMulatu T, Sintayehu Y, Dessie Y, Dheresa M. Male involvement in family planning use and associated factors among currently married men in rural Eastern Ethiopia. SAGE Open Med. 2022;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli D, Woldegiorgis AGY, Tilaye M, Yilma Y, Berhane HY, Tewahido D et al. Integrating private health facilities in government-led health systems: a case study of the public\u0026ndash;private mix approach in Ethiopia. BMC Health Serv Res. 2022;22(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMwansisya T, Mbekenga C, Isangula K, Mwasha L, Mbelwa S, Lyimo M et al. The impact of training on self-reported performance in reproductive, maternal, and newborn health service delivery among healthcare workers in Tanzania: a baseline- and endline-survey. Reprod Health. 2022;19(1).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHaddad LB, Townsend JW, Sitruk-Ware R. Contraceptive Technologies: Looking Ahead to New Approaches to Increase Options for Family Planning. Clin Obstet Gynecol. 2021;64(3):435\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilumbwe A, Nkole T, Munakampe MN, Cordero JP, Milford C, Zulu JM, et al. Facilitating community participation in family planning and contraceptive services provision and uptake: Community and health provider perspectives. Volume 17. Reproductive Health. BioMed Central Ltd; 2020.\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":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-public-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pubh","sideBox":"Learn more about [BMC Public Health](http://bmcpublichealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pubh/default.aspx","title":"BMC Public Health","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Machine learning, long-acting reversible contraception, family planning, Ethiopia, PMA survey, decision tree, health equity","lastPublishedDoi":"10.21203/rs.3.rs-6417320/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6417320/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eIntroduction\u003c/strong\u003e: Ethiopia faces challenges in Long-Acting Reversible Family Planning (LARFP) adoption despite its efficacy. Traditional statistical methods have a limited capacity to capture nonlinear determinants. This study leverages machine learning (ML) to identify predictors of LARFP use using the 2021-2023 PMA Ethiopia dataset.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e: A nationally representative sample of 9,763 women aged 15–49 was analyzed. Twenty-four variables across geographic, socioeconomic, healthcare access, and behavioral domains were preprocessed (handling missing values, encoding, and normalization). Seven ML models (Decision Tree, XGBoost, Random Forest, Logistic Regression, SVM, KNN, Naive Bayes) were trained and evaluated via stratified 5-fold cross-validation. Performance metrics included accuracy, precision, recall, F1 score, and AUC-ROC.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e: Decision Tree outperformed other models (accuracy: 99.45%, F1: 99.55%), identifying method duration (importance=0.35), provider type (0.25), and region (0.15) as top predictors. Regional disparities were stark (SNNP: 30.59% LARFP use vs. Amhara: 15.58%). Key reasons for method choice included fewer side effects (32.3%) and long duration (15.5%).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion\u003c/strong\u003e: Tree-based ML models effectively captured complex determinants of LARFP use. Targeted interventions addressing regional disparities, provider training, and client-centered care (e.g., reducing side effects) are critical for improving uptake.\u003c/p\u003e","manuscriptTitle":"Exploring Machine Learning Insights into Long-Acting Reversible Family Planning Usage in Ethiopia: Analysis of the PMA (2021-2023) Dataset","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-30 13:11:22","doi":"10.21203/rs.3.rs-6417320/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-10-15T04:39:24+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-14T16:34:21+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"309736763792294343919132680231476550276","date":"2025-09-26T18:54:44+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"325456288315652697469512890481733813138","date":"2025-09-11T15:51:30+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-10T19:23:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"284148395685339048522410801797423611080","date":"2025-05-08T19:19:16+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"109374261058696045721246127033343597711","date":"2025-05-08T11:44:42+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"178481211980833661710496682829257754583","date":"2025-05-07T16:48:44+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-29T06:33:06+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-04-11T11:10:58+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-04-11T00:32:28+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-11T00:29:45+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Public Health","date":"2025-04-10T06:54:37+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-public-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"pubh","sideBox":"Learn more about [BMC Public Health](http://bmcpublichealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/pubh/default.aspx","title":"BMC Public Health","twitterHandle":"@BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1257c326-fd37-428d-88b2-965de22849f5","owner":[],"postedDate":"April 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-01-12T16:04:40+00:00","versionOfRecord":{"articleIdentity":"rs-6417320","link":"https://doi.org/10.1186/s12889-025-26158-7","journal":{"identity":"bmc-public-health","isVorOnly":false,"title":"BMC Public Health"},"publishedOn":"2026-01-10 15:58:32","publishedOnDateReadable":"January 10th, 2026"},"versionCreatedAt":"2025-04-30 13:11:22","video":"","vorDoi":"10.1186/s12889-025-26158-7","vorDoiUrl":"https://doi.org/10.1186/s12889-025-26158-7","workflowStages":[]},"version":"v1","identity":"rs-6417320","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6417320","identity":"rs-6417320","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}