A novel and efficient multi-scale feature extraction method for EEG classification

Electroencephalography (EEG) is essential for diagnosing neurological disorders such as epilepsy. This paper introduces a novel approach that employs the Allen-Cahn (AC) energy function for the extraction of nonlinear features. Drawing on the concept of multifractals, this method facilitates the acquisition of features across multi-scale. Features extracted by our method are combined with a support vector machine (SVM) to create the AC-SVM classifier. By incorporating additional measures such as Kolmogorov complexity, Shannon entropy, and Higuchi's Hurst exponent, we further developed the AC-MC-SVM classifier. Both classifiers demonstrate excellent performance in classifying epilepsy conditions. The AC-SVM classifier achieves 89.97% accuracy, 94.17% sensitivity, and 89.95% specificity, while the AC-MC-SVM reaches 97.19%, 97.96%, and 94.61%, respectively. Furthermore, our proposed method significantly reduces computational costs and demonstrates substantial potential as a tool for analyzing medical signals.