Key predictors of maternal mild depression and anxiety in low resource settings: A machine learning approach

Background: Maternal mental health (MMH) disorders, particularly depression and anxiety, are major public health concerns in low- and middle-income countries (LMICs). In sub-Saharan Africa (SSA), their prevalence remains alarmingly high, yet identifying key predictors is challenging due to the limitations of traditional statistical methods in capturing complex risk factor interactions. Objective: This study aimed to identify key predictors of maternal depression and anxiety using machine learning (ML) techniques in a low-resource setting. Additionally, the study sought to determine the prevalence of both conditions within the population. Methods: A cross-sectional study was conducted in Kaloleni and Rabai sub-counties in Kilifi, Kenya. Data were collected from 1,995 mothers of singleton children aged 0 to 6 months between March 2023 and March 2024. Depression and anxiety symptoms were assessed using the Patient Health Questionnaire-9 (PHQ-9) and Generalized Anxiety Disorder-7 (GAD-7) scales, respectively. Additional data on sociodemographic factors, food insecurity, health history, nutrition, and socioeconomic status were collected as potential predictive features. Three supervised ML models including Random Forest (RF), Logistic Regression (LR) with L2 regularisation (Ridge), and Extreme Gradient Boosting (XGBoost) were applied to predict depression and anxiety symptoms. Model performance was evaluated using metrics including the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. Feature selection and model interpretability were performed using SHapley Additive exPlanations (SHAP) values. Results: The prevalence of maternal depression and anxiety symptoms was 15.14% and 8.67%, respectively. Although there were no statistically significant differences in prediction performance among the three models, the RF model showed a slightly better performance in predicting anxiety symptoms, with an AUC of 78.9%, accuracy of 72.9%, sensitivity of 74.3%, and specificity of 72.8%. LR performed slightly better in predicting depression symptoms, achieving an AUC of 72.4%, accuracy of 69.9%, sensitivity of 63.3%, and specificity of 71.1%. Food insecurity emerged as a key predictor for both outcomes, followed by low wealth index, increased maternal age, lower body mass index (BMI), higher number of children, and pregnancy complications. Conclusion: This study highlights the key predictors of maternal depression and anxiety in a low-resource setting, with food insecurity emerging as a critical predictor. Other key predictors included low wealth index, increased maternal age, lower BMI, higher number of children, and pregnancy complications. ML models showed potential for identifying high-risk individuals and supporting targeted interventions. However, limitations such as imbalanced data, reporting bias, and the cross-sectional design may affect generalizability. Larger, longitudinal studies are needed to validate these findings and enhance predictive performance.