Interpretable Machine Learning for Predicting Early Mental Health Care-Seeking Among Reproductive-Age Women in Bangladesh Using BDHS 2022 Data

Machine learning (ML) holds promise for predicting complex health behaviors such as mental health care-seeking, yet its use and interpretation in low-resource settings remain underexplored. This study utilized the 2022 Bangladesh Demographic and Health Survey to predict early anxiety and depression care-seeking among reproductive-age women in Bangladesh using multiple ML algorithms. We analyzed data from 4,255 ever-married women aged 15–49 with anxiety (GAD-7 ≥ 6) or depression (PHQ-9 ≥ 10). Nine ML algorithms; Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors (KNN), Gradient Boosting, XGBoost, LightGBM, CatBoost, and AdaBoost were evaluated using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUROC). The Synthetic Minority Oversampling Technique (SMOTE) was applied to address class imbalance, and SHapley Additive exPlanations (SHAP) were used to interpret the best-performing model. The Random Forest model achieved the best predictive performance (accuracy = 0.66, recall = 0.69, F1-score = 0.67, AUROC = 0.70), followed by CatBoost and Gradient Boosting (F1 ≈ 0.66). SHAP analysis identified administrative division, age group, number of children, household size, and mass media exposure as the most influential features. Geographic disparities and information access were found to have the highest SHAP values, highlighting their strong role in predicting care-seeking behavior. This study demonstrates that explainable ML, combined with SHAP interpretability, provides a reliable, transparent framework for predicting mental health care-seeking behavior in low-resource settings. These insights can inform equitable, evidence-based interventions to target at-risk populations in Bangladesh and similar contexts.