MotionXpert: EMG-Based Classification for Optimized Lower-Limb Motion Detection

Advances in surface electromyography (sEMG) signals have demonstrated significant promise for controlling exoskeletons used in gait analysis and rehabilitation. However, little is known about the optimal sEMG placement and feature selection for lower-limb classification, particularly for classifying standing and sitting movements. This study investigates the effectiveness of different sEMG electrode placements and feature sets in classifying the actions of standing and sitting using machine learning (ML) algorithms. We collected data from 22 subjects and applied noise reduction techniques before employing k-nearest neighbors classifiers, random forest, and support vector machine. Our findings show that the random forest model, using a combination of time-domain and frequency-domain features, achieved the highest F1-score (87.80 ± 5.11 %). Furthermore, the results indicate that the use of the random forest with EMG signals from the rectus femoris (RF) muscle yielded the highest classification performance in terms of F1-score (82.67 ± 8.05 %). This finding demonstrates the potential of using optimal configurations in the placement of EMG electrodes and feature sets for developing future exoskeleton-based rehabilitation applications.