Machine Learning Approaches to Predicting Depression in University Students: A Comparative Analysis of Logistic Regression, LASSO, and Random Forest

Abstract

Depression is a growing mental health concern among university students and can harm academic performance and well-being. This study analyzes data from 27,901 university students to explore key risk factors and build models to predict depression. We focus on variables such as academic pressure, financial stress, sleep duration, study satisfaction, family history of mental illness, age, gender, and grade point average. Three machine-learning methods—logistic regression, LASSO, and random forest—were applied and compared. The results show that academic pressure, financial stress, and short sleep are the strongest predictors of depression, with family history and low study satisfaction also playing important roles, while age, gender, and grades had smaller effects. All three models performed strongly, with area-under-the-curve (AUC) values around 0.86 and showed good calibration. Logistic regression and LASSO achieved nearly the same accuracy as random forest, making them easier to use and explain in real settings. These findings highlight practical steps for schools and health services: programs that lower academic pressure, provide financial help, and improve sleep habits can reduce depression risk. Early identification and timely support may help prevent serious problems. This research shows that simple machine-learning models can guide effective and affordable mental-health strategies for students.