Abstract
Wearable devices allow passive monitoring of stress-related behaviour in university students, but classifiers built on these data are not widely used in clinical practice because their decisions are not interpretable. In this paper we apply SHAP (SHapley Additive exPlanations) with TreeExplainer to an XGBoost stress classifier trained on Fitbit data and BFI-10 personality scores from 86 university students in Qatar and interpret its predictions on a held-out test set (n=61). The model reaches AUC-ROC 0.640 (95% CI 0.48-0.78), accuracy 0.672 (0.56-0.79) and F1 0.593 (0.41-0.74) on 38 features. The two features with the highest mean absolute SHAP value are exam-period context (0.156) and morning step count five days before assessment (0.154). Conscientiousness contributes to an independent personality signal. The interpretation indicates that the classifier relies on a small set of behavioural signals detectable five days before self-reported stress, suggesting a potential early-alert window for university counselling systems.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/AMAI_029.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/profile?id=%7ESarah_Aziz1
BibTex
@InProceedings{AziSar_What_MICCAISAT2026,
author = { Aziz, Sarah AND Aboubakr, Nada AND Ahmed, Arfan},
title = { { What Does an AI Model Learn About Stress? A SHAP-Based Behavioural Interpretation of a Wearable Stress Classifier } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17273},
month = {pending},
page = {pending}
}