Robust Exercise-Based Telerehabilitation for Elderly Healthcare Survives

Improving motor function post-injury or surgery often relies on physical therapy, but long waiting lists can impede effective monitoring of patient progress. Tele-rehabilitation addresses this issue by providing equitable access to therapy without the need for travel or costly in-home treatments. This paper introduces a machine learning system capable of autonomously detecting exercise correctness during patient performance, thereby enhancing rehabilitation outcomes. Utilizing the IRDS dataset, which includes 29 participants performing nine exercises, extensive feature engineering was conducted, and five machine learning algorithms coupled with various dimensionality reduction techniques were evaluated such as Random Forests (RF), K-Nearest Neighbors (KNN), Logistic Regression (LR), Support Vector Machines (SVM) and Decision Trees (DT). Random Forests combined with correlation-based feature selection emerged as the most effective model with 98% accuracy in exercise recognition, 91% accuracy in exercise correctness detection, and a combined accuracy of 95%.