Classification of sEMG Biomedical Signals for Upper-Limb Rehabilitation Using the Random Forest Method

Abstract

To use surface electromyography (sEMG) signals for therapy and rehabilitation purposes, we first need to tackle a fundamental problem which is the pattern recognition of these signals. Recently, Machine Learning (ML) techniques have drawn a lot of attention from researchers working on sEMG pattern recognition, and the usage of these techniques showed a lot of potentials and proved to be a viable option. For this work, we adopt the random forest classifier, as an ML technique, for the classification of the sEMG signals for the rehabilitation of upper limbs. Furthermore, to be able to test its performance, we considered and tested different combinations of five different time-domain features, namely MAV, WL, ZC, SSC, and finally RMS. Thus, and via experimental results on the adopted dataset, we show how the choice of features influences the quality of classification.