A Resource-Efficient Load Cell-Based Smart Bed System for Posture Classification

Smart bed systems are widely used in healthcare for continuous posture monitoring and pressure ulcer preven-tion. Existing methods, such as vision-based and biosignal-based approaches, suffer from privacy concerns, and sensor discomfort. This study proposes a contact-free, minimally instrumented load cell-based smart bed system for posture classification, aiming to provide an efficient, cost-effective alternative to high-resolution pressure sensor arrays. Our load cells were installed on the bed legs to measure the pressure distribution. A machine learning-based classification model utilizing random forest (RF), support vector machine (SVM), and k-nearest neighbor (kNN) was trained to recognize six predefined postures. User-specific data (height and weight) were integrated to enhance classification accuracy. A five-fold cross-validation strategy was applied in an inter-participant setting to ensure model generalization. The RF model achieved the highest classification performance with 80.49% accuracy, 80.49% sensitivity, 82.09% specificity, and an f1-score of 81.24%, outperforming both SVM and kNN. The confusion matrix analysis revealed that semi-lateral postures exhibited higher misclassification rates, likely due to their similar weight distribution. The proposed non-contact smart bed framework provides accurate, computationally efficient, and privacy-preserving posture classification, making it a viable alternative to high-resolution sensor arrays. Future work will focus on expanding dataset diversity, incorporating dynamic posture transitions, and integrating Internet of things-based real-time monitoring.