A comprehensive analysis of support vector machine and Gaussian mixture model for classification of epilepsy from EEG signals

Abstract

Epilepsy is one of the chronic brain disorders which affect the entire lifestyle of the person. It is characterized by recurrent seizures which are nothing but short episodes of involuntary activity or movement. For the diagnosis of epilepsy, detecting seizures is a vital step. In the clinical contexts, to detect the seizures, Electroencephalography (EEG) is used by means of visual scanning. The main aim of the paper is to reduce the dimensions of the EEG signals as the recordings of the EEG are too long to process and then classified with various post classifiers and to provide a performance comparison is provided. The dimensions of EEG signal are reduced with five different techniques like Fuzzy Mutual Information (FMI), Independent Component Analysis (ICA), Linear Graph Embedding (LGE), Linear Discriminant Analysis (LDA) and Variational Bayesian Matrix Factorization (VBMF). The dimensionally reduced values are then fed inside the Gaussian Mixture Model (GMM) and Support Vector Machines with various Kernels like Linear Kernel, Gaussian Kernel and Polynomial Kernel to classify the epilepsy from EEG signals. An exhaustive analysis is done and the results are presented with performance metrics like performance index, sensitivity, specificity, time delay, quality values and accuracy. The best result is obtained with an accuracy of 97.84% when FMI is used as a dimensionality reduction technique and followed by the usage of GMM as the post classifier.