Robustified principal component analysis for feature selection in EEG signal classification

Abstract

Feature engineering is an important step in data analysis especially for machine learning applications. Wide range of feature selection methods are being used in EEG signal processing applications. Principal Component Analysis (PCA) is considered as an ideal method for feature selection whenever high dimensional data obtained. PCA is a method to identify patterns in data streams and reproduce them in a way to highlight their similarities with unique features. The proposed work exhibits how PCA is robustified for an Electroencephalography (EEG) signal processing application by applying kernel functions. Statistical features are extracted from EEG data after preprocessing by Desecrate Wavelet Transform (DWT). Initially PCA algorithm is applied for feature selection by reducing the dimensionality. Later the algorithm is robustified by applying Gaussian kernel in a nonlinear high dimensional feature space. The research work reports that the robustified PCA produce 0.7% elevated performance than traditional PCA in an EEG classification of epileptic seizure detection.