Single channel sEMG muscle fatigue prediction: An implementation using least square support vector machine

Abstract

Surface electromyogram (sEMG) signal is commonly used for muscle fatigue analysis in clinical rehabilitation studies. Prediction results based on sEMG signals are promising because muscle contradiction can be easily characterized using sEMG signals. However, the prediction results usually deteriorate significantly when noise exist during data acquisition. Noise happens due to many factors ranging from hardware, software to procedure flaws. This investigation is aimed to assess the performance of the Least Square SVM model in predicting muscle fatigue using single channel sEMG signal. The root mean square, median frequency, and mean frequency features were extracted from two sets of raw sEMG signals captured at the multifidus (for low back pain) and flexor carpi radialis (for forearm muscle fatigue) muscles. The proposed LS-SVM technique were used to build the prediction rule-base separately for both the datasets. The implementation, testing and verification were performed in Matlab environment. The k-nearest neighbour and artificial neural network were used as the benchmarking techniques in results comparison and analysis. LS-SVM technique is proven good against the benchmarking techniques on classification accuracy and area under ROC curve. The ANOVA and Tukey HSD post hoc test were used to further validate the significant of the comparison results on both accuracy and AUC measurements.