University Students in Times of Crisis: Machine Learning-Driven Exploration of Individual and Social Determinants of Psychological Distress

The COVID-19 pandemic has significantly impacted the mental health of university students, exacerbating pre-existing vulnerabilities. This study investigates psychological distress within the Université du Québec network by examining individual and social determinants using both inferential statistics and machine learning approaches. A dataset comprising 4,691 students was analyzed using validated instruments, including the PHQ-4 for psychological distress, along with measures of spiritual well-being and social support. After preselecting key variables through bivariate analyses and controlling for multicollinearity, several predictive models were tested. Among them, CatBoost regression achieved the best performance (R2 = 0.5913, relative MAE $= 11.90\%$) in predicting PHQ-4 scores. SHapley Additive exPlanations (SHAP) were used to interpret the model’s predictions, identifying daily stress, mental health stability, and spiritual health as the most influential features. Results highlight the predominant role of modifiable, experience, based factors, such as stress perception and emotional resilience, over fixed demographic attributes. This dual-method approach offers valuable insights for designing targeted mental health interventions, especially in crisis contexts.