Integrating Machine Learning and Wearable Technology for Real-Time Physiotherapy Assistance

Purpose: This study explored the integration of wearable-based human activity recognition (HAR) systems with machine learning algorithms to enhance personalized physiotherapy monitoring.The objective was to utilize wearable sensor data for realtime posture assessment and rehabilitation assistance.Methods: Accelerometer and gyroscope data from the UCI-HAR dataset were preprocessed through standardization and dimensionality reduction using Principal Component Analysis (PCA) to improve computational efficiency.Machine learning models-Random Forest, Gradient Boosting, AdaBoost, and Support Vector Machine (SVM)-were trained and evaluated using accuracy, precision, recall, and F1score metrics.Additional analyses included confusion matrices, ROC curves, precision-recall curves, and feature importance ranking. Results:The SVM model achieved the highest accuracy (91.62%) and superior performance across most activity categories, particularly Laying (F1 = 0.98).Gradient Boosting (accuracy = 89.58%)and Random Forest (accuracy = 87.28%)also demonstrated strong predictive capability, whereas AdaBoost yielded comparatively lower accuracy (71.46%).Classification performance varied by activity type, with Walking Downstairs and Sitting showing significantly reduced scores (p = 0.03).Feature importance analysis identified accelerometer readings across the X, Y, and Z axes as the most influential variables.Conclusion: Integrating wearable HAR systems with optimized machine learning models enables accurate and real-time physiotherapy monitoring.Such systems can support clinicians in customizing posture correction and mobility training, facilitating improved patient outcomes and continuous home-based rehabilitation.