Machine Learning for Inclusive Online Mental Health Insights

Depression and anxiety impose substantial health and economic burdens yet remain under-detected and unevenly addressed in health systems and public policy. This paper presents an explainable machine learning framework for analysing depression- and anxiety-related social media discourse as a potential decision-support tool for digital governance of mental health. Using a Reddit corpus, the study combines psycholinguistic features and contextual embeddings in a hybrid BiLSTM–XGBoost ensemble, complemented by SHAP and LIME explanations and hierarchical topic modelling. The framework achieves 84.48% accuracy with balanced F1-scores, while explainability analyses clarify model predictions. Thematic patterns show how distress is linked to economic strain, barriers to care, recovery challenges, coping and peer support. These findings suggest such insights may inform public health planning, and awareness campaigns, provided wider use is embedded in ethical safeguards that prioritise accountability, data protection, and digital inclusion. Because this analysis uses Reddit data, the results should be interpreted cautiously.