Development of a hand movement classification system based on EMG signals using machine learning

Abstract

People with upper limb amputations face significant challenges in performing daily tasks, such as grasping objects or manipulating tools, which directly impacts their quality of life. Myoelectric prostheses offer a promising solution by restoring part of the lost motor function. However, controlling these prostheses remains challenging due to the complexity of electromyographic (EMG) signals and user variability. This study develops a hand movement classification system, focusing on low-cost prosthetics, using EMG signals and machine learning algorithms. The system collects EMG signals through the eHealth Sensor Platform V2.0, with electrodes placed on the forearm to ensure accurate data capture. After preprocessing, including filtering to remove noise, key features such as MAV, RMS, ZC, and SSC are extracted to capture essential information on signal amplitude and variation. A Random Forest classifier is used to identify hand open and closed movements, achieving 92.19% accuracy in tests conducted. Although the system has not yet been integrated into a prosthetic device, future steps include expanding the dataset and eventually implementing it in a low-cost myoelectric prosthesis controlled by Arduino. This research contributes to the advancement of assistive technologies, promoting accessible and effective solutions.