An HCI-Centered Posture Scoring and Feedback System with Machine Learning Regression for Personalized Sitting Behavior Monitoring Using Smart E-Textile Sensing

This study presents a human-centred sitting monitoring system that moves beyond discrete posture classification by estimating a continuous Healthy Sitting Score (HSS, 0–100) from electro-textile sensing data. Forty-eight tactile features were used with PINNR, SVR, and GPR models, and generalization was evaluated using participant-wise Group Cross-Validation and an independent unseen-subject test set. The best PINNR model achieved R = 0.948, R² = 0.898, and RMSE = 6.83 on unseen participants. The system was integrated with a Bluetooth-enabled Android application providing real-time HSS visualization, history, summaries, and configurable notifications. In a 21-participant workplace evaluation, usability ratings were high and HSS increased significantly after the first notification, although the absence of a no-feedback control limits causal interpretation. Repeated-use analysis over 21 workdays showed no widespread sensor drift. Overall, the study emphasizes continuous posture-quality scoring, personalized feedback, behavioural response, and real-world HCI evaluation rather than multi-class posture recognition.