A Comparative Analysis of Machine Learning Models for Classifying College Student Mental Health Status Using Psychological Factors

Abstract

College student mental health is a growing concern globally, necessitating effective and scalable methods for early identification and intervention. Machine learning (ML) offers promising avenues for analyzing complex psychological data to predict mental health status. This study conducts a comprehensive comparative analysis of various ML algorithms for classifying the mental health status of college students based on psychological assessment factor scores. Data derived from psychological assessments, encompassing ten factor scores (Somatization, Obsessive-Compulsive, Interpersonal Sensitivity, Depression, Anxiety, Hostility, Phobic Anxiety, Paranoid Ideation, Psychoticism, Other) and a total score, were preprocessed using MinMax scaling. Seven distinct ML models were implemented and evaluated: Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), Random Forest (RF), Logistic Regression (LogReg), Support Vector Machine (SVM), XGBoost, and a hard Voting ensemble classifier combining the predictions of the individual models. A robust evaluation framework using 10-fold stratified cross-validation was employed to assess model performance based on Accuracy, Precision, Recall, F1-Score, and Area Under the Receiver Operating Characteristic Curve (AUC). The results demonstrate the high predictive potential of the selected features, with most models achieving strong performance. Notably, ensemble tree-based methods, XGBoost (Accuracy: 0.992, AUC: 1.00) and Random Forest (Accuracy: 0.992, AUC: 1.00), exhibited superior performance and stability across cross-validation folds. The Voting classifier also showed robust performance (Accuracy: 0.965), although its reported AUC calculation presented anomalies requiring cautious interpretation. Other models like Logistic Regression (Accuracy: 0.956, AUC: 1.00) and MLP (Accuracy: 0.953, AUC: 1.00) performed well, while SVM (Accuracy: 0.942, AUC: 1.00) and LSTM (Accuracy: 0.934, AUC: 0.99) showed comparatively lower, yet still respectable, results for this dataset. The findings underscore the efficacy of ML, particularly ensemble techniques, in classifying student mental health status based on standardized psychological factor scores, suggesting potential for integration into campus mental health screening and support systems. Further research should focus on validating these models on diverse populations, incorporating additional data types, and exploring model interpretability.