Comparison of Elbow Angle Classification Performance Using sEMG-Based Random Forest, SVM, and XGBoost Models

sEMG 신호를 이용한 Random Forest, SVM, XGBoost 기반 팔꿈치 각도 분류 성능 비교

This study aimed to develop a machine learning model for classifying elbow angles using sEMG signals. Previous research in rehabilitation and robotic arm control has often combined sEMG and IMU sensors to measure muscle activity and precise angles. However, in rehabilitation, utilizing sEMG signals alone may be more practical than using multiple sensors. The use of sensors like IMUs increases equipment costs and complicates data processing, making interpretation more challenging. In contrast, sEMG signals reflect muscle activation and can predict angles simply and effectively, making them suitable for assessing elbow movements. This study classified elbow flexion and extension angles into 15°, 30°, 45°, 60°, and 90° using only sEMG sensors. sEMG data were collected from the biceps brachii and triceps brachii muscles and analyzed using Random Forest, SVM, and XGBoost models to evaluate angle classification performance. The experimental results showed high accuracy for all three models, with SVM and XGBoost demonstrating particularly superior performance. These findings suggest that sEMG signals alone can effectively predict elbow angles in applications such as rehabilitation and arm control, providing a valuable tool for assessing and aiding recovery of motor functions in rehabilitation therapy.