Efficient Early Prediction of Hand Movements Using Two-Channel Electromyography Signal Analysis

The electrical activity produced by muscles is recorded by the electromyography signal, a dynamic biosignal. This dynamic nature represents complex muscular behaviour patterns. Electromyography -based recognition systems have far-reaching effects, revolutionizing human-computer interactions, enabling sign language detection, and giving amputees greater control over their devices.This study aims to predict specific hand movements by analyzing electromyography signals from a minimal number of channels. The computational cost efficiency of this approach makes it suitable for real-time applications. Electromyography signals were collected from the forearm muscles of 15 healthy subjects (5 women and 10 men) aged between 19 and 24. Preprocessing tasks were applied to the raw signals, including high-pass filtering, Butterworth filtering, notch filtering, and signal rectification. The dataset consists of recordings from four channels, of which two channels were selected for predicting hand movements. Two 500 ms time windows from both channels were used as inputs for a weighted K-Nearest Neighbors algorithm. The task involved predicting and classifying the intentions of three hand movements: pinching, finger abduction, and grasping. The procedure's overall results demonstrate 83.1% detection accuracy for the weighted K-Nearest Neighbors indicating high precision and a relatively short response time for predicting these hand movements in healthy subjects.