Comparison of Machine Learning Algorithms for EMG based Muscle Function Analysis

Electromyography (EMG) is the process of measuring the electrical activity that is produced during the muscle activation and is also called as myoelectric signal. This electrical activity can be recorded to analyze the performance of any muscle. Machine learning algorithms can be deployed to predict the presence of abnormality in the muscles. In this study, we have worked on predicting the possibilities of cervicalgia, more commonly known as neck pain, by using distinct time domain features extracted from surface EMG signals recorded from SCM muscle, and performing the comparative analysis of different machine learning binary classifier algorithms like Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest, Decision Tree, Naïve Bayes Classifier and Logistic Regression. The feature selection was made using techniques like PCA and Pearson's Correlation Matrix. Maximum classification accuracy of 91.6% was achieved. If predicted well in advance, they can provide valid information to doctors, physiotherapists, yoga therapists or medical practitioners, who can then adapt their diagnosis and treatment for each patient.