Machine Learning for Teacher Depression Prediction: A Systematic Literature Review of Risk Factors, Predictive Models, Explainability, and Deployment

Gilbert Yegon, Edward Ombui, Collins Oduor

Depression among teachers is a global public health concern with significant consequences for educator well-being, teaching effectiveness, and student outcomes. Despite growing evidence regarding its prevalence and associated risk factors, the application of machine learning (ML) for predicting and explaining teacher depression remains relatively underexplored. This systematic literature review synthesizes current evidence on the use of ML techniques to predict teacher depression, with particular emphasis on the sociodemographic, occupational, and clinical risk factors used as predictive features, the ML algorithms employed and their comparative performance, the explainability techniques supporting model interpretation and decision-making, and the challenges and opportunities for real-world deployment. A comprehensive literature search was conducted across PubMed, Scopus, Web of Science, IEEE Xplore, and the ACM Digital Library in accordance with the PRISMA guidelines. Eligible studies included those applying supervised or unsupervised ML approaches to model depression risk among teachers, with extracted data covering study design, geographical context, feature sets, algorithms, performance metrics, explainability methods, and deployment considerations. The review found that logistic regression, random forests, support vector machines, and gradient boosting were the most frequently adopted algorithms, while key predictive factors included sociodemographic characteristics (e.g., gender, marital status, and age), occupational stressors (e.g., workload, grade level, and learner-to-teacher ratio), and clinical history, particularly previous psychiatric diagnoses. Among explainable artificial intelligence (XAI) techniques, SHAP and LIME were the most widely used to enhance model transparency and support decision-making. However, evidence of practical deployment, especially in low-resource settings and African contexts, remains limited. Overall, the review demonstrates that ML has considerable potential to facilitate the early identification of teachers at risk of depression, although its broader application is constrained by limited high-quality datasets, insufficient model explainability, and inadequate consideration of equity across gender, race, and geographical settings. These findings provide a comprehensive evidence base to inform future research, guide the development of robust and interpretable predictive models, and support policy translation and implementation in educational systems.