Electro-textile smart chair cover for multi-class office sitting posture recognition: subject-independent and cross-chair validation with mobile IoMT feedback

Abstract

Prolonged sitting in office environments is a complex, human-centered behavior shaped by inter-subject variability (e.g., BMI/anthropometry), real-world uncertainty caused by furniture changes and sensing variability. This study proposes an internet of medical things (IoMT) sitting posture monitoring system that integrates a multi-surface electro-textile smart chair cover, automatic ground-truth generation, and deployment-oriented mobile analytics for decision support. Five sitting posture classes were defined based on medical recommendations and relevant standards. Participants maintained each posture for 30 s on two different office furniture types equipped with the smart cover, yielding two datasets. To minimize labeling subjectivity, skeletal joint coordinates were recorded synchronously via a depth sensor and posture-specific joint angles were computed to automatically label electro-textile measurements. A comprehensive benchmark was performed using convolutional neural network (CNN), long short-term memory (LSTM), support vector machine, k-nearest neighbors, Naive Bayes, discriminant analysis, decision tree, and ensemble learning. To quantify generalization under controlled cross-chair variability, the evaluation included subject-independent testing (participant-level separation) and cross-dataset transfer between furniture settings (A → B and B → A) without re-tuning on the target dataset. The best-performing models achieved near-ceiling performance, with 1D-CNN and LSTM reaching 99.8% and 99.7% overall accuracy; in transfer tests, 1D-CNN maintained 98.2% (A → B) and 96.5% (B → A), indicating promising transferability under controlled office-chair domain shift. Finally, a Kotlin/TFLite mobile application was developed to compute posture scores and deliver real-time feedback; usability tests suggested improved user awareness and positive perception. Overall, the proposed framework links sensing, intelligent analytics, and mobile decision support to monitor standard seated behavior in everyday workplaces.