EMG scalogram-based classification of gait disorders using attention-based CNN: a comparative study of wavelet functions

Abstract

This study aims to classify gait abnormalities caused by rheumatoid arthritis and prolapsed intervertebral disc using scalograms from the EMG signals. Classifying EMG signals is difficult because of their variability, high dimensionality, and sensor placement. We propose to bridge this gap by using the wavelet transform and attention-based neural networks. The study involved five participants: one with rheumatoid arthritis, two with prolapsed intervertebral disc, and two healthy subjects. The proposed methodology uses four different wavelet functions: complex Gaussian, frequency B Spline, Mexican Hat, and Shannon, to construct scalograms, and an attention-based CNN for classification. A comparison of performance of the proposed algorithm with nine machine learning classifiers: K nearest neighbour, Naïve Bayes, support vector machine, decision tree, logistic regression, random forest, AdaBoost, gradient boost, and XGBoost was conducted. Out of the nine machine learning classifiers that were tested, XGBoost achieved the highest accuracy of 90.38%, however, in comparison to this the performance of the proposed algorithm was much better, with an accuracy of 99% and precision of 99%. These results indicate that this approach is highly effective in accurately categorising EMG signals. Copyright © 2024 Inderscience Enterprises Ltd.