EMG Data Generation for Scooter Riding and Classification Using Machine Learning to Control a Prosthetic Arm

This paper presents a novel EMG-based control method for a prosthetic limb aimed at scooter riders, targeting brake and accelerator functions. EMG signals were acquired from the flexor muscles of healthy subjects using surface electrodes while simulating scooter control movements. A six-feature time-domain feature set was extracted, and classification was performed using SVM, Random Forest, KNN, and a 10-layer Artificial Neural Network (ANN). The ANN model achieved the highest classification accuracy of 87.5%. The processed signal was integrated into a control logic simulating actuation for prosthetic accelerator/brake operations. This task-specific adaptation distinguishes the study from conventional EMG-based prosthetics.