Classification and Feature Extraction of Different Hand Movements from the EMG Signal using Machine Leaning based Algorithms

Abstract

Prosthetic plays an important role for the amputees to improve the ability and mobility of their regular activities. Electromyography(EMG) has been used for decades in the control of the motorized upper-limb prosthesis. Processed EMG can imitate human movements. Mayo armband is a wireless sensor of low power, Bluetooth, and small interference which provides a good quality EMG signal. The Myo armband measures the EMG from the upper limb. In this thesis work, the statistical time-domain features have been considered to classify different hand movements. The classification and comparison have been performed by 4 different Machine Learning-based algorithms i.e. Support Vector Machine(SVM), Naïve Bayes(NB), Random Forest(RF), and K-Nearest Neighbor(KNN). The data has been collected from subjects (males and females) of different ages. The classifier model has used 80% data as a training set and the remaining 20% of data as the test set. The result shows that Random Forest and SVM outperform the other two algorithms with an accuracy of 98%. Referring to the accuracy here, this classification model serves as a promising candidate for the input of prosthetic hand control systems.