Prediction of Mental Health Problem Using Annual Student Health Survey: Machine Learning Approach

Area of Science:

• Machine Learning Applications in Mental Health
• Educational Psychology
• Public Health Surveillance

Background:

• Student mental health support often relies on self-reported surveys, but standardized criteria for intervention are lacking.
• Proactive identification of students at risk is crucial for timely mental health support.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting students' mental health problems one to two years in advance.
• To assess the utility of health survey content and response time data in mental health prediction.

Main Methods:

• Utilized health survey data from 3561 undergraduate students (2020-2021).
• Compared various ML models (logistic regression, random forest, XGBoost, LightGBM) for prediction accuracy.
• Analyzed the impact of demographic data, survey responses, and answering time on mental health predictions.

Main Results:

• The LightGBM model demonstrated high predictive performance in both one-year and two-year predictions.
• Responses related to campus life, anxiety, and future outlook were the most significant predictors.
• Answering time variables offered minimal additional predictive improvement, though some were influential.

Conclusions:

• Health survey data, including demographic and behavioral information, can effectively predict future mental health issues.
• The developed ML model integrates health survey characteristics with ML capabilities for improved student mental health monitoring.
• Findings can inform enhancements to health survey design and student outreach protocols.