Understanding student mental health: An explainable pattern analysis approach

Student mental health is a growing global concern, with stress, nervousness, and related psychological challenges negatively affecting academic performance and overall well-being. Existing studies often rely on isolated variables or black-box predictive models that lack interpretability. This study aims to identify key risk factors for student mental health through explainable pattern analysis, enabling actionable insights for early intervention. Three student mental health datasets encompassing demographic, behavioural, and psychosocial attributes were analyzed using a decision tree algorithm. The model extracted interpretable patterns linking coping difficulty, anger over poor performance, perceived loss of control, and difficulty relaxing to reported stress and nervousness levels. Patterns were assessed using support (number of records) and confidence (likelihood of outcome) to determine the most influential predictors. Results showed that coping difficulty emerged as the most informative feature in the decision tree model, showing the strongest statistical association with nervousness and stress, with confidence levels reaching 99.15%. Anger due to poor performance and difficulty relaxing further increased stress likelihood, even among students with minimal coping struggles. Interestingly, stress appeared in students without coping difficulties (71.56% confidence), suggesting the presence of hidden or unmeasured influences such as financial strain or social isolation. The decision tree produced interpretable, rule-based pathways that consistently identified similar high-risk profiles, demonstrating reliability and generalizability across three datasets. This study provides novel insights into how psychosocial factors collectively shape student mental health outcomes. The findings underscore the value of explainable AI for developing data-driven, proactive support systems in educational settings and highlight the need for future integration of longitudinal and multimodal data.