Comparative Study Between Machine learning algorithms and feature ranking techniques on UI-PRMD dataset

Abstract

Rehabilitation exercises reduce the demand for healthcare services over time by decreasing the number of hospital visits, lengths of stay, and readmissions. Since rehabilitation is a continuous process, it is crucial to monitor patient progress. This paper compares various machine learning classifiers which enable patients to perform exercises at home instead of visiting a physiotherapy center. The system assesses the correct performance of the exercises and tracks the patient's improvement, leading to lower rehabilitation costs. A distinct skeletal part, angle, and trajectory are required for each activity to distinguish between the workouts and assess whether they were executed correctly. Data extraction was performed using one Kinect camera, and six feature ranking algorithms were employed to construct the system, with the top features selected. Subsequently, 13 classical machine learning algorithms were implemented to identify the algorithm that produced the most accurate classification results. According to our experiments, Extra Tree Classifier, which employs feature extraction using the ReliefF technique, produces the best classification results, with an accuracy score of 99.94%.