Improved Differential Evolution with Stacked Auto Encoder for EEG Motor Imagery Classification

Abstract

Objectives: To develop an improved version of Differential Evolution (DE) algorithm to overcome the complexity in extracting the features from the Electroencephalogram (EEG) based Brain-Computer Interfaces (BCI) systems; To develop a Stacked Auto Encoder (SAE) for classifying motor imagery signals into left, right, feet and tongue movements, respectively. Methods: Improved Differential Evolution Optimization Algorithm (IDEOA) is proposed for the selection of features which is extracted by the hybrid CSP-CNN feature extraction model. Extracted features will undergo the classification process by using SAE. Findings: The proposed IDEOA has an accuracy of 97.34% compared to the existing Sinc-based convolutional neural networks that obtained 75.39% and TSGL-EEG-Net of 81.34%. Novelty: The proposed IDEOA improves the mutation strategy results in improved convergence effect.