Hand Movement Classification from SEMG Signals Using Machine Learning

Trans-radial amputation is one of the most common upper limb amputations, which occurs below the elbow, proximal to the wrist. Hence, trans-radial prosthetics like myoelectric prosthetics are a common choice for the upper limb. Myoelectric prosthetics use electrical signals generated by the muscle contractions, which are then detected by the surface electrodes to control and facilitate the movements of the limb. However, amputees struggle to control these prosthetics, and hence, machine learning is used to help minimize the time required for amputees to operate their new prosthetic hand and improve the quality of their life. This can be achieved by training the machine learning models in better hand movement recognition to analyze and correctly predict the user's intent. In this paper, we explored the above by training our model with XGBoost on sEMG data, which consisted of six hand movements performed by ten healthy individuals, including both men and women, to develop movement classification that could serve as a foundation for future prosthetic applications. To benchmark the performance of XGBoost, we compared it against SVM, Random Forest, and KNN classifiers, which achieved accuracies of 72.12 %, 84.40 % and 82.42 % respectively, while XGBoost achieved 83.34 %. Although this study uses data from healthy individuals, proper validation on the amputee population is required for real-world implementation in the field of prosthetics.