Multi‑source Data Prediction of College Students’ Psychological Stress and Model Interpretability Testing
Abstract Aiming at the drawbacks that the assessment of college students’ psychological stress relies on single scales and suffers poor timeliness, this study constructs a multi‑source feature set integrating questionnaire scales, wearable physiological signals and campus behaviour records. Six models covering three categories (ridge regression, logistic regression, decision tree, random forest, gradient boosting decision tree and Stacking ensemble) are compared for prediction performance. Stratified five‑fold cross‑validation and nested hyper‑parameter search are adopted, with macro‑average F1‑score, AUC and Brier score as evaluation metrics. The results show that multi‑source fusion improves the macro‑average F1‑score by approximately 8‑12 percentage points compared with single‑source data. Gradient boosting decision tree and Stacking ensemble achieve optimal performance. SHAP and permutation importance analyses indicate that sleep‑duration fluctuation, resting heart‑rate variability and proportion of nocturnal activity rank highest in feature contribution. Ablation experiments verify the complementary gains of physiological and behavioural features for prediction. All codes, random seeds and data‑partition schemes are reproducible.
Authors
- Dong Li (ORCID: https://orcid.org/0009-0000-0188-2269)
- Qiang Ma
Institutions
- Fuyang Normal University (CN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-13
- DOI
- https://doi.org/10.21203/rs.3.rs-11304486/v1
- Primary Topic
- Artificial Intelligence in Healthcare
- Type
- preprint