Predicting Positive Psychological States using Machine Learning and Digital Biomarkers from wearable data

Wearable devices provide continuous physiological data, offering opportunities for real-world mental health monitoring. While prior research has primarily focused on detecting stress and negative psychological states, the prediction of positive psychological states remains underexplored. This study investigates whether digital biomarkers derived from smartwatch data can predict daily psychological states based on the PERMA framework (Positive Emotion, Engagement, Relationships, Meaning, and Accomplishment). Using data from 34 participants wearing research-grade smartwatches for eight days, physi-ological signals—including heart rate variability (HRV), electrodermal activity (EDA), and accelerometer-derived movement features—were processed into machine-learning-ready features. Machine learning models, including Random Forest, Long Short-Term Memory networks (LSTM), Convolutional Neural Networks (CNNs), and Transformers, were trained to predict daily self-reported psychological states. Explainable AI techniques, such as SHAP and LIME, were employed to identify key physiological markers, while conformal prediction assessed model uncertainty. Results indicate that while absolute prediction accuracy remains a challenge, CNN models achieved a threshold accuracy of up to 62%, with accelerometer-based features showing the highest predictive importance. Self-esteem-related constructs demonstrated the most distinguishable physiological patterns, while anger was more detectable than anxiety or sadness. Uncertainty analysis further highlighted the potential for improving confidence in predictions. These findings suggest that wearable-derived biomarkers hold promise for tracking posi-tive psychological states, albeit with limitations in predictive accuracy. Future work should focus on refining feature extraction, improving model generalization, and integrating multi-modal data sources to enhance real-world applicability in mental health monitoring.