Performance of modified power spectral density features in EEG signal classification

Abstract

This paper evaluates the performance of classification of Electroencephalogram (EEG) data by focusing on several normalization and dimensionality reduction processes in Power Spectral Density (PSD) signal pre-processing. It focuses on effect of modification of PSD features as an input for classification of EEG signals. For ANN classification, Zero-mean normalization method produces the best performance when compared against other complicated dimensionality reduction techniques such as Locally Linear Embedding (LLE) and Orthogonal Least Squares (OLS). The improvement achieved by Zero-mean normalization in ANN is 4.5% better than Baseline PSD. For SVM classification, PCA produces best performance with an enhancement as much as 10% better than Baseline PSD. It found that SVM classifier performs significantly better than ANN classifier in classifying variants of PSD features.