The comparison of various dimension reduction and classification methods in EEG signals

In this study, it is aimed to find out the impact of features derived from epileptic and non epileptic EEG signals and the reduction of their dimensions by Principal Component Analysis (PCA) and Independent Component Analysis (ICA) on classification success and to compare the classification performances of Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM). A total of 20 people,10 epilepsy patients who were diagnosed by specialist physicians and 10 non-epilepsy patients were included in the study. The EEG recordings taken from those who were diagnosed with epilepsy while they were not having seizures. To classify epileptic and non epileptic signals, first and foremost Discrete Wavelet Transform (DWT) and the spectral analysis of EEG signals were performed and features that would be used to classify the signals were obtained. First, features were classified without their dimentions being reduced. Then, they were classified after their dimensions were reduced by PCA and ICA. In this way, both the effects of PCA and ICA on the classification performance and the classification performances of LDA and SVM in which linear and radial basis kernel functions are used, were determined. While dimension reduction methods PCA and ICA improved classification performances of LDA and SVM in which linear basis function is used, they decreased the classification performance of SVM in which radial basis function is used. Also similar classification results were obtained by features whose dimensions were reduced by PCA and ICA. Generally calassification performance of SVM in which radial basis function is used, was higher than the other classification methods and the highest classification success with 92.5% sensitivity, 85.6% specificity, and 88.9% accuracy ratios were obtained by SVM method in which features, whose dimensions were not reduced and radial basis kernel function is used. Using the features obtained from the DWT coefficients a good discrimination can be obtained between normal and epileptic signals. Classification performance depends on using EEG signals, the features obtained from these signals, dimension reduction methods and the classification methods. Finally due to high classification performance of SVM in which radial basis function is used, it can be used as a decision support tool for physicians to diagnose epilepsy.