Authors:
Ricardo Henrique Avelar Matheus
and
Maria Claudia F. Castro
Affiliation:
Department of Electrical Engineering, Centro Universitário FEI, Brazil
Keyword(s):
sEMG, Machine Learning, Feature Extraction, LDA, RMS, Amputees.
Abstract:
One of the most significant challenges to the quality of life for amputees is the development of prostheses that can closely simulate the capabilities of the lost limb. One possible solution to this problem is myoelectric prostheses, which are devices that use myolectric signals as users’ intention to perform independent movements. This study aims to investigate how optimizing feature extraction methods can impact the performance of machine learning models in recognizing surface electromyogram (sEMG) signals from amputees. The LibEMG library in Python, which offers a simple and robust API for developing sEMG-based projects, was used alongside the DB8 dataset from the NINAPRO public database, which promotes machine-learning research in human, robotic, and prosthetic hands. A total of twelve feature extraction methods and seven different classifiers were tested. The results showed the best mean accuracy of 79.18% using a Random Forest classifier with a set of eleven time and frequency
domain features, considering the data of an amputee with experience in using myoelectric prostheses. However, the most affected models by feature optimization were KNN, MLP, and SVM, with accuracy improvements up to 69.28%.
(More)