Multivariate EEG Signal Using PCA and CNN in Post-Stroke Classification

Abstract

A stroke occurs due to circulatory disorders in the brain that can cause severe disability. Therefore, it requires rehabilitation. One of the instruments used in monitoring post-stroke patients is the Electroencephalogram (EEG). However, EEG signals generated from multiple channels often experience data redundancy, affecting the computational time load and accuracy. Therefore, it needs to reduce the dimensions of the data of the EEG signal. This research proposed a model to classify post-stroke patients based on EEG signals that used Wavelet, Principal Component Analysis (PCA), and Convolutional Neural Networks (CNN). Wavelet transform is used to extract EEG signals into Delta, Alpha, Theta, and Mu waves. PCA works to remove some signal of multi-channel by selecting the number of components. Meanwhile, classification used one dimension Convolutional Neural Network (CNN). Experimental results gave using PCA with 45 components produced an accuracy of 93.33% compared without using PCA, which results in an accuracy of 86.66%. Besides, optimization models using AdaDelta provided higher accuracy compared to Adam optimization models.