Quantifying Academic Risk Factors for Student Depression Using WHO Frameworks: Odds Ratios, SHAP Explainability, and Tipping Point Analysis
- 1University of Barishal
Abstract
Depression has become a serious concern for students worldwide. Aligned with the WHO Helping Adolescents Thrive (HAT) Guidelines and the Social Determinants of Health (SDoH) model, this study isolates five academically relevant factors- academic pressure, work/study hours, study satisfaction, sleep duration, and financial stress- from a dataset of 27,880 university students in India and quantifies their associations with depression. Unlike prior work that maximises classification accuracy, this study prioritises interpretability: logistic regression provides odds ratios (OR) with 95% confidence intervals, Random Forest (RF) and XGBoost rank predictors by feature importance, and SHAP (SHapley Additive exPlanations) values extend the analysis to individual-level risk explanation. Sleep Duration is encoded as an ordinal integer (0-3) preserving its natural order, and Spearman correlations are computed on the original 27,880-row dataset prior to any resampling. SMOTE oversampling was applied exclusively to the training set, and performance was evaluated on the original imbalanced test set (n = 5,576). Both ensemble models achieve approximately 77-78% accuracy and an AUC of 0.845, confirmed by 5-fold pipeline cross-validation (CV AUC [~] 0.843). Academic pressure is the dominant risk factor (OR = 2.271; RF importance = 0.481; mean |SHAP| = 0.174), while study satisfaction (OR = 0.796) and sleep duration (OR = 0.835) are protective. Interaction analysis reveals how depression risk is amplified by low sleep, high financial stress, and extended study hours: under average conditions, the 50% risk threshold is crossed at Academic Pressure {approx} 3.04, and under compounded adverse conditions, this tipping point shifts to even lower pressure values. These findings provide data-driven thresholds aligned with WHO-endorsed modifiable determinants to support early detection and institutional counselling.
Related
- Linking academic stress to loneliness, anxiety, and depression: The association of problematic micro-drama use among Chinese university students
- Suicide risk factors in nursing students: Depressive symptoms, academic burnout and academic efficacy after the pandemic
- Depression and anxiety, and associated factors among female community health volunteers in Parsa district, Nepal
- Perceived Stress and Associated Factors among Undergraduate Nursing Students during Academic Clinicals: An Analytical Cross-Sectional Study
- Depression and associated factors among women of childbearing age : A cross-sectional study based on the health ecology model
BibTeX
@misc{ahmed2026quantifying,
title = {{Quantifying Academic Risk Factors for Student Depression Using WHO Frameworks: Odds Ratios, SHAP Explainability, and Tipping Point Analysis}},
author = {T. Ahmed and M. R. A. Asif},
howpublished = {medRxiv},
year = {2026},
month = aug,
doi = {10.64898/2026.08.09.26360019},
url = {https://www.medrxiv.org/content/10.64898/2026.08.09.26360019v2},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Quantifying Academic Risk Factors for Student Depression Using WHO Frameworks: Odds Ratios, SHAP Explainability, and Tipping Point Analysis
AU - T. Ahmed
AU - M. R. A. Asif
PY - 2026
DA - 2026/08/09
PB - medRxiv
DO - 10.64898/2026.08.09.26360019
UR - https://www.medrxiv.org/content/10.64898/2026.08.09.26360019v2
AB - Depression has become a serious concern for students worldwide. Aligned with the WHO Helping Adolescents Thrive (HAT) Guidelines and the Social Determinants of Health (SDoH) model, this study isolates five academically relevant factors- academic pressure, work/study hours, study satisfaction, sleep duration, and financial stress- from a dataset of 27,880 university students in India and quantifies their associations with depression. Unlike prior work that maximises classification accuracy, this study prioritises interpretability: logistic regression provides odds ratios (OR) with 95% confidence intervals, Random Forest (RF) and XGBoost rank predictors by feature importance, and SHAP (SHapley Additive exPlanations) values extend the analysis to individual-level risk explanation. Sleep Duration is encoded as an ordinal integer (0-3) preserving its natural order, and Spearman correlations are computed on the original 27,880-row dataset prior to any resampling. SMOTE oversampling was applied exclusively to the training set, and performance was evaluated on the original imbalanced test set (n = 5,576). Both ensemble models achieve approximately 77-78% accuracy and an AUC of 0.845, confirmed by 5-fold pipeline cross-validation (CV AUC [~] 0.843). Academic pressure is the dominant risk factor (OR = 2.271; RF importance = 0.481; mean |SHAP| = 0.174), while study satisfaction (OR = 0.796) and sleep duration (OR = 0.835) are protective. Interaction analysis reveals how depression risk is amplified by low sleep, high financial stress, and extended study hours: under average conditions, the 50% risk threshold is crossed at Academic Pressure {approx} 3.04, and under compounded adverse conditions, this tipping point shifts to even lower pressure values. These findings provide data-driven thresholds aligned with WHO-endorsed modifiable determinants to support early detection and institutional counselling.
ER -