Predicting depression among public university students in Bangladesh using machine learning models

Mental health challenges, particularly depression, anxiety, and stress, are prevalent among university students worldwide. In Bangladesh, public university students face unique stressors, including financial pressure, academic demands, social isolation, and limited access to mental health services. This study aims to identify key risk factors for depression and develop a predictive model to facilitate early intervention. A cross-sectional study was conducted across 15 public universities in Bangladesh, involving 500 students selected through random sampling. Data were collected via an online survey on socio-demographic, academic, and psychological factors. Depression was assessed using a structured questionnaire developed based on PHQ-9 items and aligned with DSM-5 criteria. Descriptive statistics and chi-square tests were used to explore the distribution and associated factors. Machine learning models (Random Forest, SVM, KNN, and Logistic Regression) were applied. Independent t-tests were performed on precision, recall, accuracy, F1-score, and specificity to determine the best model. The prevalence of depression (85.5%) was higher among students, however, there was no statistical significance (p = 0.095) among male students (87.8%) compared to females (81.9%), though. Lower CGPA (p < 0.001), financial stress (p < 0.001), lack of family support (p < 0.001), exposure to ragging (p < 0.001), and frequent session jams (p < 0.001) were significantly associated with depression. Because of precision, recall, accuracy, F1 score and Specificity; the SVM is best model for the dataset to predict depression. Based on SHAP analysis consultancy service, educational system and session jam were found to be the most influential parameters and addressing them can be a game changer. Depression among public university students in Bangladesh are influenced by academic, financial, and institutional factors. Machine learning models, especially SVM offer promising tools for early depression risk prediction, enabling targeted interventions to improve student well-being.