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

Area of Science:

• Computational social science
• Mental health informatics
• Machine learning applications

Background:

• Early identification of student psychological issues is critical for timely support and campus safety.
• Distinguishing normal behavior from distress signals in complex campus data is challenging.
• This study addresses the need for sensitive methods to detect abnormal behavior and predict psychological health.

Purpose of the Study:

• To develop and evaluate a sensitive temporal modeling approach for detecting abnormal behavioral patterns.
• To predict psychological health states in students using high-dimensional campus data.
• To improve the early identification of students requiring psychological support.

Main Methods:

• A two-phase methodology involving Jenks natural breaks for feature discretization and Apriori for association rule mining.
• Development of an attention-enhanced gated module to model historical behavioral time series.
• Integration of long-term habits and short-term fluctuations with dynamic weighting of irregular behavioral changes.

Main Results:

• The proposed method significantly outperformed baseline models on the StudentLife dataset across key metrics.
• The attention-enhanced gated module effectively captured temporal mental health state evolution.
• Prioritization of discriminative behavioral anomalies proved effective in prediction.

Conclusions:

• Combining association rule mining with attention-based temporal modeling is effective for psychological health prediction.
• The approach provides a practical tool for campus administrators to identify at-risk students.
• Proactive intervention is enabled through the analysis of complex behavioral data.