[Paper Review] AdaptStress: Online Adaptive Learning for Interpretable and Personalized Stress Prediction Using Multivariate and Sparse Physiological Signals

Xueyi Wang, Lamoth, Claudine J. C.+1 more|arXiv (Cornell University)|Feb 19, 2026

Digital Mental Health Interventions0 citations

TL;DR

AdaptStress develops an explainable, personalized stress forecasting model using multivariate wearable signals with domain adaptation and selective test-time adaptation to predict stress days ahead across 16 participants.

ABSTRACT

Continuous stress forecasting could potentially contribute to lifestyle interventions. This paper presents a novel, explainable, and individualized approach for stress prediction using physiological data from consumer-grade smartwatches. We develop a time series forecasting model that leverages multivariate features, including heart rate variability, activity patterns, and sleep metrics, to predict stress levels across 16 temporal horizons (History window: 3, 5, 7, 9 days; forecasting window: 1, 3, 5, 7 days). Our evaluation involves 16 participants monitored for 10-15 weeks. We evaluate our approach across 16 participants, comparing against state-of-the-art time series models (Informer, TimesNet, PatchTST) and traditional baselines (CNN, LSTM, CNN-LSTM) across multiple temporal horizons. Our model achieved performance with an MSE of 0.053, MAE of 0.190, and RMSE of 0.226 in optimal settings (5-day input, 1-day prediction). A comparison with the baseline models shows that our model outperforms TimesNet, PatchTST, CNN-LSTM, LSTM, and CNN under all conditions, representing improvements of 36.9%, 25.5%, and 21.5% over the best baseline. According to the explanability analysis, sleep metrics are the most dominant and consistent stress predictors (importance: 1.1, consistency: 0.9-1.0), while activity features exhibit high inter-participant variability (0.1-0.2). Most notably, the model captures individual-specific patterns where identical features can have opposing effects across users, validating its personalization capabilities. These findings establish that consumer wearables, combined with adaptive and interpretable deep learning, can deliver relevant stress assessment adapted to individual physiological responses, providing a foundation for scalable, continuous, explainable mental health monitoring in real-world settings.

Motivation & Objective

• Address the need for individualized, interpretable stress forecasting using consumer wearables.
• Develop a multivariate, sparse,time-series framework that handles missing data and captures cross-variable interactions.
• Incorporate domain adaptation and test-time adaptation to mitigate inter-participant variability and enable deployment across diverse users.
• Provide explainability analyses to identify dominant predictors and support clinical trust.

Proposed method

• Use a Transformer-based encoder to learn 15-dimensional multivariate physiological time-series representations.
• Incorporate a feature-level attention mechanism to dynamically weight signals for stress prediction.
• Apply adversarial domain adaptation with a domain discriminator to learn domain-invariant representations across participants.
• Implement selective Test-Time Adaptation (TTA) to decide when adaptation is beneficial for a new participant, using entropy minimization and pseudo-labeling.
• Process data with an overlapping sliding window (history window and forecast window) to generate multiple training samples and forecasts.
• Use Leave-One-Out cross-validation across 16 participants and evaluate with MSE, MAE, RMSE, Pearson correlation, and trend direction accuracy, plus SHAP for explainability.

Experimental results

Research questions

• RQ1 Can a multivariate, sparse representation of wearable-derived signals predict future stress for individuals with varying baselines?
• RQ2 How well does domain adaptation reduce cross-participant predictive gaps in stress forecasting?
• RQ3 Does selective Test-Time Adaptation improve predictions for new participants without harming those already well-aligned?
• RQ4 Which physiological features most strongly and consistently explain stress across individuals?
• RQ5 Do personalized patterns emerge where the same features have different effects across users?

Key findings

• The model achieved MSE 0.053, MAE 0.190, RMSE 0.226 under optimal settings (5-day input, 1-day prediction).
• AdaptStress outperformed TimesNet, PatchTST, CNN-LSTM, LSTM, and CNN across all conditions, with improvements up to 36.9% over the best baseline.
• Sleep metrics emerged as the most dominant and consistent stress predictors (importance 1.1, consistency 0.9–1.0), while activity features showed high inter-participant variability (0.1–0.2).
• The model captures individual-specific patterns where identical features can have opposing effects across users, validating personalization capabilities.
• Leave-One-Out evaluation demonstrates cross-participant generalization when combined with domain adaptation and selective TTA.