Globally, depression has seen a significant increase of 25% in the context of the COVID-19 pandemic. This study investigates the socioeconomic, biological, and social factors associated with depression in Peru. Previous Peruvian research primarily relied on cross-sectional analyses with traditional regression techniques, limiting the ability to detect complex, nonlinear interactions between determinants. Addressing these gaps, this study applies explainable machine learning methods–including Random Forest, Gradient Boosting, and XGBoost–alongside interpretability tools like SHAP to identify and rank key predictors of depression severity over multiple years. Model performance was robust, with XGBoost achieving high accuracy (82%) and AUC-ROC (0.89). These findings underscore the importance of integrated mental and physical health care interventions, social protection, and gender-sensitive strategies tailored to Peru’s socio-demographic landscape. The study demonstrates the value of machine learning to provide transparent, actionable insights for targeted public health policies aimed at reducing depression’s burden. Its multi-year, interpretable approach advances understanding beyond prior analyses, emphasizing disability and socioeconomic determinants as urgent priorities to improve mental health outcomes in Peru. This research offers significant contributions to evidence-based mental health policy formulation in a middle-income country context.