Feature and muscle selection for an effective hand motion classifier based on electromyography

Abstract

An issue that arises in the hand motion classification based on the electromyography (EMG) system is the failure of choosing the right features and number of muscles. These parameters are fundamental in determining the accuracy and effectiveness of the classifier system. Therefore, the objective of this study is to develop and evaluate an effective hand motion classifier based on the EMG signal. The three-channel of EMG was collected by placing three pairs of electrodes on the surface of the skin. Six statistic features (mean, variance, standard deviation, kurtosis, skewness, and entropy) were selected to extract the EMG signal using a window length of 100 samples. A muscle and features selection is applied to the classifier machine (linear discriminant analysis (LDA), support vector machine (SVM) and K nearest neighborhood (KNN)) to retrieve the most useful feature and muscle. In this study, we found that there was no significant difference in accuracy among a number of muscles (p-value>0.05). LDA and SVM showed the best accuracy and no significant difference in accuracy between both were found. This study concluded that EMG signal from a single muscle can classify the hand motion (hand close, open, wrist flexion, and extension) effectively.