Classification and Feature Extraction of Electroencephalogram Signals for Epilepsy Using PCA, ICA, DWT and SVM Methods

The purpose of this article is to classify electroencephalogram signals into two types of epilepsy and healthy. To achieve the highest accuracy, various techniques have been used. The desired characteristics of these signals can be extracted by Wavelet Transform and Empirical Mode Decomposition methods. These two methods are compared in terms of impact in the classification process. To reduce the dimensions of the feature space, Independent and principal Component Analysis methods can be used. Then, in order to reduce the effect of noise on electroencephalogram signal analysis, a smoothing method can be applied. Finally, by using Support Vector machine classifier, the existing data classified. These steps were tested for an existing data set, including 5 groups of single channel electroencephalogram signals. Results show that the empirical decomposition method has high efficiency and accuracy to extract the characteristics and classification of signals. Accordingly, the accuracy and sensitivity of both combinations of "empirical mode decomposition - independent component analysis" and "empirical mode decomposition - principal component analysis", after data smoothing, as a new approach to extraction and classification of features are 100%. The output of this system is used to control and treat the disease.