Investigating the Cause of the Poor Identification Rate in SVM Classification Using EMG During Movement in the Target Direction

Abstract

We have been developing a human-machine interface for hemiplegic patients. We have been estimating the direction from the EMG generated during shoulder joint motion, but the identification rate was less than 50%. In this paper, we investigated the cause of the low identification rate based on the data length, the number of data, and the number of features. As a result, the number of features influenced identification. However, there were some who had an influence on the identification rate with respect to the data length and the number of data and others who did not.