Efficacy of Dimensionality Reduction Techniques and 2D Features in Classification of EEG Data

Abstract

EEG-based BCIs enable a direct path for the human brain to communicate with external devices. The main objective of BCI is to classify brain signals to control some applications of interest. EEG is a paradigm that helps to obtain the brain signals from the different states of the human brain. EEG signal feature extraction is a critical topic in brain signal processing that has attracted great attention nowadays with the emergence of large-scale data. The focus of this paper is to reduce the dimension and redundancy of raw EEG data to facilitate retrieval of selected dynamic information. Three approaches of feature extraction and dimensionality reduction viz. PCA, LDA and ICA have been implemented on EEG datasets recorded during attention and meditation state of the brain. Moreover, as massive amount of information about the relationship between time and frequency coefficients is mostly missing in EEG dataset. hence, 1D feature set has been transformed into 2D images. The efficacy of the proposed methods has been tested through the CNN classifier for the classification of EEG signals. The experimental results show that the proposed method is able to achieve an average classification accuracy of 100%.