Deep Learning based classification of motor imagery EEG signals using an improved path finder optimization algorithm

Abstract

Motor Imagery Brain-Computer Interfaces (MI-BCIs) are systems based on AI that collect patterns of brain activities in mental movement and translate these movements through external devices. The identification of motor intention by evaluating Electroencephalogram (EEG) signals is an important issue in applications related to Brain-Computer Interfaces (BCI). In this paper, an Improved Path Finder Optimization Algorithm (IPFOA) is proposed to improve the process of selecting features. The data is collected from the BCI competition IV 2a dataset and the stage of pre-processing takes place through sliding windows and 6th order Butterworth filter. The feature will be extracted from the pre-processed data using the hybrid Convolutional Neural Network (CNN) and Common Spatial Pattern (CSP) method. After this stage, the required features are selected from the extracted features using the proposed IPFOA algorithm. Finally, the classification of EEG signals takes place using a Stacked autoencoder, a classifier based on a Deep Neural Network. The experimental results show that the proposed approach achieved a better accuracy of 98.40% which is comparatively higher than the existing approaches.