Employing Neighborhood Component Analysis as a Feature Selection Method in the Pattern Recognition Approach for Surface Electromyography Signal Classification

Many research works regarding the design of a pattern recognition (PR) approach for surface electromyography (sEMG) signal analysis have attempted to find the most relevant features leading to the highest classification performance. While some classical works have outlined the general time-domain (TD) features related to sEMG signals, the research area still lacks data-driven feature selection (FS) methods that would generalize the effect of features on the classification performance. This paper proposes a novel approach of using a Neighborhood Component Analysis (NCA) algorithm as the FS step in the PR scheme. The NCA-based feature selection (NCA-FS) has resulted in an increase in the classification accuracy and F1 score of four machine learning (ML) models while reducing the total complexity of the PR scheme for two subjects in the NinaPro Database 2 (DB2). Moreover, the proposed method has elicited the most powerful TD features, positively affecting the overall performance, compared to some classical and state-of-the-art works regarding sEMG signal recognition.