An interpretable machine learning framework for student depression risk prediction: integrating multi-domain features with SHAP-based explainability

Background Depression among university students constitutes a pressing global mental health challenge, with prevalence rates exceeding 30% in numerous populations. While traditional screening instruments such as the Patient Health Questionnaire-9 (PHQ-9) offer effective clinical detection, they are inherently reactive and focus solely on symptom assessment, lacking the ability to capture modifiable risk factors. Machine learning (ML) approaches, particularly ensemble methods with explainability frameworks, present a promising avenue for risk stratification using multi-domain data. Objective This study aimed to develop and validate an interpretable XGBoost-based prediction model integrating academic, demographic, lifestyle, and clinical features to predict student depression risk, and to systematically evaluate model performance, feature contributions, and clinical utility using SHAP (SHapley Additive exPlanations) and Decision Curve Analysis (DCA). Methods A cross-sectional dataset of 27,901 student records with 16 multi-domain features was analyzed. XGBoost was trained with 80/20 stratified split (n_train = 22,320, n_test = 5,581). Model performance was assessed via area under the receiver operating characteristic curve (AUC), calibration (Brier score), and DCA. Feature importance was quantified through SHAP values, and eight predefined hypotheses concerning risk factor directionality, dose-response relationships, and domain contributions were tested. Results The XGBoost model achieved AUC = 0.919, accuracy = 0.845, F1-score = 0.869, and Brier score = 0.111. The strongest predictors by mean | SHAP| were suicidal ideation (1.212), Academic Pressure (0.934), and Financial Stress (0.679). Financial Stress and Work/Study Hours demonstrated clear monotonic dose-response relationships. CGPA showed significant linear association (R 2 = 0.473, p < 0.001). The multi-domain model significantly outperformed all single-domain models ( p < 0.001). DCA confirmed net clinical benefit across threshold probabilities of 10%–50%. A sensitivity model excluding suicidal ideation maintained good performance (AUC = 0.868), supporting its practical utility in ethically constrained screening contexts. Conclusion The XGBoost + SHAP framework demonstrates robust predictive performance, excellent calibration, and clinically meaningful interpretability for student depression risk assessment. Multi-domain integration is essential for comprehensive risk profiling. These findings support the development of explainable ML-based screening tools for educational mental health settings.

Authors

Institutions

Publication Details

Journal
Frontiers in Psychology
Published
2026-11-18
DOI
https://doi.org/10.3389/fpsyg.2026.1936485
Primary Topic
Artificial Intelligence in Healthcare
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

An interpretable machine learning framework for student depression risk prediction: integrating multi-domain features with SHAP-based explainability

Xue Chang
3 links
Frontiers in Psychology
Artificial Intelligence in Healthcare
article

An interpretable machine learning framework for student depression risk prediction: integrating multi-domain features with SHAP-based explainability

Xue Chang
article en

Abstract

Background Depression among university students constitutes a pressing global mental health challenge, with prevalence rates exceeding 30% in numerous populations. While traditional screening instruments such as the Patient Health Questionnaire-9 (PHQ-9) offer effective clinical detection, they are inherently reactive and focus solely on symptom assessment, lacking the ability to capture modifiable risk factors. Machine learning (ML) approaches, particularly ensemble methods with explainability frameworks, present a promising avenue for risk stratification using multi-domain data. Objective This study aimed to develop and validate an interpretable XGBoost-based prediction model integrating academic, demographic, lifestyle, and clinical features to predict student depression risk, and to systematically evaluate model performance, feature contributions, and clinical utility using SHAP (SHapley Additive exPlanations) and Decision Curve Analysis (DCA). Methods A cross-sectional dataset of 27,901 student records with 16 multi-domain features was analyzed. XGBoost was trained with 80/20 stratified split (n_train = 22,320, n_test = 5,581). Model performance was assessed via area under the receiver operating characteristic curve (AUC), calibration (Brier score), and DCA. Feature importance was quantified through SHAP values, and eight predefined hypotheses concerning risk factor directionality, dose-response relationships, and domain contributions were tested. Results The XGBoost model achieved AUC = 0.919, accuracy = 0.845, F1-score = 0.869, and Brier score = 0.111. The strongest predictors by mean | SHAP| were suicidal ideation (1.212), Academic Pressure (0.934), and Financial Stress (0.679). Financial Stress and Work/Study Hours demonstrated clear monotonic dose-response relationships. CGPA showed significant linear association (R 2 = 0.473, p < 0.001). The multi-domain model significantly outperformed all single-domain models ( p < 0.001). DCA confirmed net clinical benefit across threshold probabilities of 10%–50%. A sensitivity model excluding suicidal ideation maintained good performance (AUC = 0.868), supporting its practical utility in ethically constrained screening contexts. Conclusion The XGBoost + SHAP framework demonstrates robust predictive performance, excellent calibration, and clinically meaningful interpretability for student depression risk assessment. Multi-domain integration is essential for comprehensive risk profiling. These findings support the development of explainable ML-based screening tools for educational mental health settings.

Frontiers in PsychologyVol. 17
University of Nottingham Ningbo China (CN), Shenzhen City Polytechnic (CN)
Good health and well-being
Artificial Intelligence in Healthcare
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon