Comparative Analysis of Classifiers for EMG Signals

Abstract

Electromyographic signals have a considerable importance in robotic hand prosthesis and various biomedical applications. The analysis of these signals for pattern recognition of arm movements is helpful to facilitate the handicap individuals with upper limb impairment or paralysed individuals who are able to reinstitute innate control of hand. These signals need to be recorded from the patients hand with the help of electrodes which may be contaminated with noises or undesired signals which affects the output accuracy. To ensure the detection of data with reduced noise and to execute the optimal performance from the analysis, the signals are preprocessed. The data collected for the 52 movements from 27 different subjects is provided by NinaPro database which allows the whole research community to add more advancement to this field. The purpose of this research is to analyse the dataset from this easily accessible database for twelve finger and hand movements acquired from 27 subjects. This processed data was then tested for two different classifiers, Linear Discriminant Analysis classifier and Artificial Neural Network classifier, to examine their percent classification accuracy. The data classified with Linear Discriminant Analysis gives the mean classification accuracy of 85.41% while Artificial Neural Network classifier shows 91.14%. The results for the dataset used in this study revealed that Artificial Neural Network performs better in the classification and recognition of data for hand movements as compared to Linear Discriminant Analysis.