sEMG-Based Knee Abnormality Detection Using EWT-Entropy-Ensemble Algorithm During Walking

An essential problem that can be identified using a surface electromyography (sEMG) signal is identifying knee abnormality. The manual analysis of EMG data takes much time and expertise from medical professionals. Thus, the proposed work is based on a diagnosis of knee abnormalities automatically during walking with features derived from sEMG signals and enhanced with empirical wavelet transform (EWT) and entropy-based features using a Bagging Ensemble Classifier. In this study, 22 people (11 with and 11 without knee abnormalities) had their lower limb (sEMG) data gathered while walking from the UCI repository. Since noise frequently tampers with sEMG, an approach for sEMG denoising based on EWT was presented. EWT decomposes the original signal into several multiresolution analyses. To reduce the computational complexity of the signal, the overlapping windowing technique was used, with a window size of 250 ms and a 50% overlap. Entropy features were extracted channelwise, and using ANOVA feature selection approaches, extremely significant entropy features that may distinguish functional knee abnormalities efficiently were retrieved from the decomposed signals. Last, machine learning models, such as support vector machine, k-nearest neighbors, naïve Bayes, and ensemble bagged tree were achieved with an accuracy of 99.6%, precision of 99.5%, recall of 99.8, F1-score of 99.7%, and with training time reduced to 5.66 s.