Boosting EMG classification: a hybrid NCA-driven evolutionary optimization approach for high accuracy and efficiency

Abstract

A novel hybrid approach combining neighborhood component analysis (NCA) and metaheuristic optimization algorithms is proposed to improve the classification accuracy of electromyography (EMG) signals while reducing the feature set size and computational time. EMG signals were collected from six neck and shoulder muscles, and a total of 23 features were extracted, including 17 time domain and 6 frequency domain features. The extracted features underwent pre-processing and selection using both filter-based and wrapper-based techniques. Among the evaluated metaheuristic algorithms, gray wolf optimization (GWO) and equilibrium optimization (EO) showed the best performance. The proposed hybrid approaches, NCA-GWO and NCA-EO, combined the strengths of NCA for feature ranking with the search efficiency of GWO and EO. These hybrid methods achieved high classification accuracy, reduced feature subsets, and improved computational efficiency compared to individual methods. Statistical validation using Wilcoxon signed-rank tests confirmed that the performance of NCA-GWO and NCA-EO was statistically comparable. Evaluation metrics such as accuracy, mean squared error (MSE), kappa coefficient, and Fisher’s score (F-score) demonstrated the robustness and reliability of the proposed methods. The results suggest that NCA-GWO and NCA-EO have significant potential for real-time EMG classification applications.