Automated EMG Signal Classification at the Epoch Level Using Wavelet Scattering Features and KNN

Abstract

Accurate and efficient classification of electromyography (EMG) signals is essential for objective assessment of neuromuscular disorders. In this study, an automated EMG classification framework was developed at the epoch level by combining wavelet scattering–based feature extraction with a K-Nearest Neighbor (KNN) classifier. EMG recordings from amyotrophic lateral sclerosis (ALS), myopathy, and healthy control subjects were processed through a MATLAB-based pipeline. The raw signals were resampled to 23,437 Hz, baseline-corrected, and band-pass filtered between 20 and 2,500 Hz to eliminate motion and power-line artifacts. Each recording was divided into one-second epochs, and the Wavelet Scattering Transform (WST) was applied to extract robust, noise-insensitive time–frequency features. The resulting feature matrices were normalized and classified using multiple supervised algorithms within MATLAB’s Classification Learner, including KNN and neural network models. Among the evaluated classifiers, the Fine KNN achieved the best overall performance, reaching a validation accuracy of 93.36% and a test accuracy of 95.98% while maintaining low computational cost. Neural network models achieved comparable accuracy but required substantially higher training time. The findings demonstrate that WST-based feature extraction combined with KNN classification provides a reliable, efficient, and reproducible approach for EMG signal analysis at the epoch level. This work underscores the potential of wavelet scattering as a compact and robust feature representation technique for biomedical signal processing applications.

Keywords: Biosignal Analysis, EMG, WST, Signal Processing, Classification