Machine Learning based Classification of Upper Arm Movements using sEMG Signal

This paper presents classification of upper arm movements by acquiring EMG signals from upper limb muscles of 9 participants. These movements were adduction, upward movement and front movement. The four channeled electrode was placed at deltoid, trapezius, biceps and triceps to acquire signals during upper arm movement. The signal preprocessing was performed by applying notch and band pass filter. The frequency of 50Hz was removed from the signal. Band pass filter allowed signals with range between 10-200Hz. Sliding window with size of 250ms was applied for partitioning data. Time domain and frequency domain features were extracted from preprocessed signal. The Willison Amplitude for signals of classes were comparatively analyzed. The classification and detection of upper arm movement was conducted using four classifiers namely Decision Tree, Random Forest, KNN and SVM. Accuracy and precision were selected as performance parameters. The system provided an accuracy of 82.7% with Decision tree and precision of 85.7% with Random Forest.