Overview
Mental health is a foundational element of student well-being, influencing academic success and social integration. A cross-sectional study conducted at two public Bangladeshi universities (Jahangirnagar University, Dhaka, and Patuakhali Science and Technology University) provides a comprehensive look at how common depression, anxiety, and stress are among university students, and how machine learning (ML) can help predict risk. By combining traditional epidemiology with advanced ML techniques, the study offers evidence to guide campus policies and targeted interventions in Bangladesh and similar LMICs.
Prevalence and severity
Among 1,697 respondents, the study found high rates of psychological distress: depression in 56.9%, anxiety in 69.5%, and stress in 33.2%. Severity varied, with substantial shares experiencing moderate to extremely severe symptoms across the three conditions. Compared with international benchmarks, these figures are notably elevated, particularly for anxiety. The results align with COVID-19-era data showing heightened distress among students and resonate with broader LMIC patterns of mental health burden in university settings.
Key associated factors
The study identified several demographic, familial, and behavioral factors linked to mental health outcomes:
- <strongGender: Female students showed higher odds of stress, while males exhibited higher depression levels in certain analyses, underscoring nuanced gender effects that likely reflect academic pressures and social expectations.
- <strongFamily relationships: Unfriendly family dynamics were robustly associated with all three outcomes (depression, anxiety, and stress), highlighting the central role of family support in Bangladeshi student well-being.
- <strongAcademic factors: Being in earlier or later years and faculty affiliation influenced risk. Notably, students in commerce reported higher depression and anxiety, possibly tied to career uncertainty and competitive job markets.
- <strongSubstance use: Cigarette smoking and illicit drug use were significantly linked to depression and anxiety, suggesting intertwined health behaviors and mental health risk.
These associations emphasize a broad spectrum of influences—from intimate family context to discipline-specific pressures—that universities must consider when designing mental health strategies.
Machine learning in predicting risk
To explore predictive potential, the researchers evaluated several ML models, including CatBoost, XGBoost, SVM, Random Forest, KNN, Gradient Boosting, and Logistic Regression. The data were split 80/20 for training/testing, with SMOTE used to address class imbalance in the training set. Models were assessed via 5-fold cross-validation using accuracy, precision, F1-score, log loss, and AUC-ROC.
Findings showed that no single model achieved strong discriminatory power (AUC-ROC values hovered around 0.5–0.6). However, some models performed better for specific outcomes: SVM excelled for depression and anxiety in terms accuracy and precision, CatBoost showed strong performance for stress, and LR demonstrated competitive results across several metrics. Feature importance analyses (SHAP values and XGBoost gains) consistently highlighted family relations, gender, and smoking status as influential predictors, with academic discipline also contributing meaningfully in certain models.
Interpretation: While ML offered valuable insights into risk patterns, predictive accuracy remains modest for these multifactorial mental health outcomes. This underscores the need for richer feature sets, possibly including longitudinal data, digital behavioral markers, and campus-specific variables, to strengthen future models.
<h2Implications for policy and practice
Given the high prevalence of depression, anxiety, and stress, Bangladeshi universities should consider routine mental health screening integrated into student services, coupled with confidential counseling and targeted support for high-risk groups (e.g., students in commerce, those reporting unfriendly family relationships, and those engaging in smoking or other substance use).
Additionally, family-informed interventions, campus-wide mental health literacy campaigns, peer-support networks, and faculty training can help foster a supportive environment. Policymakers should recognize student mental health as a priority and encourage data-informed strategies that balance privacy with proactive care. The study’s ML analyses point to the potential for digital risk assessment tools, provided ethical safeguards and stakeholder transparency are ensured.
Concluding thoughts
This Bangladeshi study adds to the global dialogue on student mental health by detailing prevalence, risk factors, and the cautious promise of ML-based prediction. Its integrative approach—blending traditional epidemiology with modern ML—offers practical avenues for intervention while acknowledging current methodological limits. As universities in LMICs expand mental health services, embracing both human-centered care and data-driven insights will be essential for safeguarding student well-being and academic success.
