Predicting Mental Health Issues in College Students Using Machine Learning

This is quantitative research that explored cause-and-effect predictions of mental health of college students based on advanced causal inference and machine learning classification methods. The study conducted a cross-sectional analysis of 101 university students, identifying depression, anxiety, and panic attacks prevalence based on the Student Mental Health dataset (Shariful07, 2020). This study used a two-fold analytic literature, wherein five causal inference approaches were used to predict the gender impacts on mental health outcomes adjusting for confounding factors, and three supervised learning algorithms were used to build predictive models. Findings indicated a prevalence rate of 34.7, 33.7, 32.7, and a high degree of comorbidity of mental conditions prevalence of depression, anxiety, and panic attacks respectively. Regression adjustment, Mantel-Haenszel stratification, direct standardization, propensity score, and instrumental variables all provided convergent estimates, and E-values showed that they were robust to unmeasured confounding. Machine learning models had a range of AUC-ROC of 0.52-0.71 with best results by XGBoost. The analysis of the importance of the feature revealed that marital status, age, and academic variables were the important predictors. This paper has shown that causal inference and machine learning are complementary in risk factor knowledge and prediction, respectively, and have implications in terms of early detection and intervention in university mental health services.