ABSTRACT
Interfaces that use sEMG signals face the challenge of correctly identifying
the signal while distinguishing it from noise or interference.
Although classical techniques like visual inspection and machine
learning methods exist, most studies focus on signals from healthy
individuals. There is a lack of data and methods suitable for signals
from individuals with neurological conditions, such as cerebral
palsy and post-stroke. This study analyzes sEMG data from individuals
with neurological injuries, using machine learning methods to
identify muscle contractions and rest without pre-processing. The
data were acquired from people with neurological diseases, such as
cerebral palsy and post-stroke. They were extracted using sEMG
from triceps brachii and extensor carpi radialis muscles. The signals
were not preprocessed and were input as segmented time windows
to three proposed classifiers: Support Vector Machine, Random Forest
and an Ensemble Voting classifier. All three classifiers reached
around 99% accuracy and F1-Score on typical sEMG data, but the
results on abnormal data were inconclusive.