Finger movement recognition using machine learning algorithms with tree-seed algorithm

Abstract

Electromyography (EMG) signals have been used to recognize various actions of hand movements, finger movements, and hand gestures. This paper aims to improve the classification accuracy of EMG signals while decreasing the number of features using the tree-seed algorithm. The dataset containing EMG signals utilized in this investigation is derived from a publicly accessible source. The rationale for selecting the tree-seed algorithm centers on its ability to enhance classification accuracy while minimizing the dimensionality of feature sets. The object function and tree-seed algorithm’s nature avoids the results to have low accuracy with fewer features. The aim is not just to use a smaller number of features but also to achieve a higher accuracy rate. To ensure that selecting a smaller number of features does not decrease classification accuracy, the performance of all feature subsets was evaluated using the objective function. As a result, the number of selected features decreased, while the accuracy rate increased. The best accuracy improvement was observed, with the rate rising from 84.78% to 90.21% using the k-nearest neighbor (kNN) classifier with 50 out of 80 features. The maximum classification accuracy achieved was 99.75%, also using the kNN classifier. In this study, two different feature sets were compared using two different optimization algorithms in conjunction with four traditional machine learning algorithms to evaluate changes in classification accuracy. The classification accuracy and the improvements in accuracy, along with the number of selected features at the end of the iterations, have been reported.