Mental health issues among college students have become a pressing concern, severely affecting both their academic performance and overall wellbeing. Predicting and understanding the mental health status of college students remains a complex task due to three main challenges: the limited availability of large-scale longitudinal datasets, the use of opaque black-box machine learning models, and the reliance on population-level analysis rather than personalized insights. In this study, we address these challenges by leveraging advanced machine learning algorithms to predict and interpret college student mental health. Specifically, we implement Random Forest, Adaboost Regressor, Gradient Boosting Regressor, and XGBoost Regressor to analyze a comprehensive dataset spanning five years, which includes data collected both before and during the COVID-19 pandemic. By using this dataset, which is the longest longitudinal mobile sensing dataset of its kind, we aim to provide accurate predictions and offer valuable insights into the factors affecting mental health among college students. This work not only focuses on improving prediction accuracy but also aims to enhance the interpretability of the models, providing more transparent and actionable insights for mental health professionals.

Keywords: College student mental health, machine learning, interpretable models, Random Forest, Adaboost Regressor, Gradient Boosting, XGBoost Regressor, mobile sensing dataset, longitudinal data, COVID-19 pandemic, mental health prediction, personalized analysis.