A method for predicting student psychological health based on behavioral time series analysis.

Background: Early identification of student psychological issues is essential for providing timely support and preventing safety incidents. However, distinguishing between normal behavioral variability and critical indicators of distress within complex, high-dimensional campus data remains a significant challenge. This study proposes a sensitive temporal modeling approach designed to detect abnormal behavioral patterns and predict psychological health states. Methods: We developed a two-phase methodology to enhance feature quality and temporal sensitivity. First, the Jenks natural breaks algorithm was employed for optimal feature discretization to manage data heterogeneity. Subsequently, the Apriori algorithm was utilized to perform association rule mining, filtering for behavioral features with the strongest correlations to specific psychological states. Finally, we implemented a gated module enhanced by attention mechanisms to model historical behavioral time series. This module was specifically designed to integrate long-term habits with short-term fluctuations, dynamically assigning higher weights to irregular behavioral changes. Results: The model was evaluated using the public StudentLife dataset. Comparative experiments demonstrate that the proposed method significantly outperforms several baseline models across four key evaluation metrics. The results indicate that the attention-enhanced gated module effectively captures the temporal evolution of mental health states by prioritizing discriminative behavioral anomalies. Conclusions: Our findings validate the efficacy of combining association rule mining with attention-based temporal modeling for psychological health prediction. This approach offers a practical, precise tool for campus administrators to identify students at risk, enabling proactive intervention through the analysis of complex behavioral data.