Feature extraction of EMG signals, classification with ANN and kNN algorithms

Abstract

This study aims to classify the electrical signals generated by the movement of muscles. The data set contains eight different motion signals from four different individuals. Classified motions are thumb, index, middle, ring, thumb-index, thumb-middle, thumb-ring and hand close. Firstly, the signal is divided into small parts by windowing process, and the feature vectors are created by applying various feature extraction methods to these small signals. Then, the feature vectors are classified by k-nearest neighbors and artificial neural networks algorithm. At the end of these processes, the classification accuracy was 89% for kNN at 150 ms and 93% for ANN. The advantage of the kNN Algorithm is that the processing time is shorter than ANN.