The escalating global prevalence of mental health disorders necessitates robust predictive frameworks to enable timely interventions and optimize resource allocation. This study leverages machine learning (ML) models linear regression (LR), random forest (RF), gradient boosting regressor (GBR), and decision tree (DT) to forecast depression and anxiety trends among youth and elderly populations across regions from 2020 to 2030. Utilizing age- stratified datasets, the research evaluates model performance through accuracy, mean absolute error (MAE), and mean squared error (MSE). For depressive disorders, GBR achieved superior accuracy (0.9984) with the lowest MSE (0.0014) and MAE (0.0285), followed closely by RF (accuracy: 0.9970, MAE: 0.0269). Conversely, LR excelled in anxiety prediction, attaining near-perfect accuracy (0.9999) with minimal errors (MAE: 0.0052, MSE: 0.00005). The analysis underscores the importance of model-dataset alignment, with ensemble methods like RF and GBR proving effective for complex depression patterns, while simpler models like LR suffice for linear anxiety trends. By integrating age-specific features and forecasting decade-long trends, this work provides actionable insights for policymakers to design targeted mental health strategies, emphasizing early detection and demographic-tailored interventions. The results highlight ML’s transformative potential in public health analytics, enabling scalable, data-driven solutions to mitigate the burden of mental health disorders.