Search for optimal features for direction estimation using SVM classification using EMG generated when moving the shoulder in a specified direction

Abstract

We are developing a Human-Machine Interface using surface Electromyogram (sEMG) for hemiplegic patients. Our approach employs machine learning to estimate the direction of shoulder movements from sEMG signals directed toward a target. In this paper, we aim to improve accuracy by using 35 features and employing a feature selection algorithm to identify highly contributive features. As a result, the accuracy with the initial 35 features was 62%. By utilizing a feature selection algorithm to select features with high contribution rates, the accuracy improved to approximately 76%.