Comparison of Machine Learning Algorithms for EMG Signal Classification

Abstract

The use of muscle activation signals in the control loop in biomechatronics systems is extremely important for effective and stable control. One of the methods used for this purpose is motion classification using electromyography (EMG) signals that reflect muscle activation. Classifying these signals with variable amplitude and frequency is a difficult process. On the other hand, EMG signal characteristics change over time depending on the person, task and duration. Various artificial intelligence-based methods are used for movement classification. One of these methods is machine learning. In this study, a total of 24 different models of 6 main machine learning algorithms were used for motion classification. With these models, 7 different wrist and static hand gestures (rest, grip, flexion, extension, radial deviation, ulnar deviation, expanded palm) are classified. Test studies were carried out with 8 channels of EMG data taken from four subjects. Classification performances were compared in terms of classification accuracy and training time parameters. According to the simulation results, the Ensemble algorithm Bagged Trees model has been shown to have the highest classification performance with an average classification accuracy of 98.55%.