Hand Gesture Classification using sEMG Signals and Ensemble Learning

Abstract

Electromyographic signals offer details on a muscle activity of a person. In the case of hand movements various combinations of forearm muscles must be activated in order to carry out each gesture, which results in unique electrical patterns. On the other hand, identifying the motion being executed is made possible by the examination of these patterns of muscle activation, which are captured by EMG signals. This article presents a novel implementation on hand gesture classification utilizing Electromyography (EMG) signals recorded through a MYO Thalmic bracelet. The dataset consists of raw data collected from 36 subjects, each executing a series of static hand gestures. The classification model employed for this study is an ensemble of decision trees, achieving an impressive validation accuracy of 98.0% and a test accuracy of 98.1 %. The article outlines the dataset details, recording methodology, and provides an in-depth analysis of the model architecture, hyperparameters, and training results.