EMG Signal Analysis and Real-Time Muscle Activation Classification using Random Forest

Electromyography (EMG) plays a crucial role in the detection and diagnosis of muscle activities. It was found that EMG signals captured from body muscles can be used for human-computer interfaces and prosthetic control. This paper presents Machine Learning based muscle state identification through classification of the EMG obtained at real-time using surface electrodes and Advancer Technologies Muscle Sensor v3. In this work, the signals have been captured from the bicep muscle and are interfaced using an Arduino UNO. The signals are filtered using low-pass filter, moving average filter and peak- to-peak measurement method prior to further processing. The system was found to achieve an accuracy of 99% using random forest, trained on an average of 3000 data values representing the bicep muscle activation and relaxation obtained from 10 individuals. A comparison has been done with the commonly used Support Vector Machine (SVM) method also. The developed classification model is further used for control of opening and closing of computer-based applications.