Impact of feature selection on sEMG signal classification

Abstract

The sEMG signal contains both relevant and irrelevant features. In order to reduce the computational burden, time, and cost of hardware development, only the selection of relevant features is necessary. This research article reports an impact analysis of feature selection on hand gesture classification based on surface electromyography (sEMG) signals. For this purpose, the analysis of variance (ANOVA) algorithm is used to rank the features. A subject selection method is developed on the basis of ranked features, to select a generalized subject. Four classifiers, including Decision Tree (DT), Support Vector Machine (SVM), k-Nearest Neighbor (kNN), and Naïve Bayes (NB), have been considered to test the impact of feature selection. The performance of classifiers before and after feature selection is compared on the basis of accuracy, precision, recall, f1-score, training, and testing time. Average accuracy and time consumption improves from 70.04% and 0.13425 seconds to 89.75% and 0.03845 seconds after ANOVA based feature selection is employed. Additionally, four channels are identified to reduce complexity of acquisition device.