Electromyography (EMG) Data-Driven Load Classification using Empirical Mode Decomposition and Feature Analysis

This study presents a novel methodology towards load classification through surface-Electromyography (sEMG). Two subjects performed weight lifting tasks for 1kg, 3kg and 7kg weights. 50 samples were collected from each subject. sEMG signals were denoised through Empirical Mode Decomposition (EMD). Region of interest extraction for improving classification performance is also achieved through EMD by removing redundant information from preprocessed signal. Feature extraction is performed through extensive analysis of different combinations of statistical, time domain, frequency domain and Mel frequency cepstral coefficient (MFCC) features. Best feature set is constructed based on highest classification accuracy. Selected features were validated using 10-fold cross validation over range of different classifiers such as Support Vector Machine (SVM), Decision Tree (DT), Quadratic Discriminant Analysis (QDA), Naïve Bayes (NB), and K-nearest neighbor. SVM with cubic kernel yields best classification accuracy of 99% for three class problem.