Conceptual Analysis of Epilepsy Classification Using Probabilistic Mixture Models

In the past two decades, the Electroencephalograph (EEG) dependent Brain Computer Interface (BCI) for analyzing and detecting the mental disorders especially epilepsy has triggered a lot of research interest in both biomedical industrial side and academia. The main ingredient of EEG dependent BCI are preprocessing of EEG signals, feature extraction of EEG signals and classification of EEG signals. Very rich and useful information about the electrical activities of the brain is provided by the EEG. The amplitude and frequency varies in the EEG signal when various mental tasks are executed. Due to the lengthy nature of the EEG data, computing it becomes quite hectic. Therefore in this paper, the dimensions of the lengthy EEG recorded data is reduced with the help of Principal Component Analysis (PCA), Expectation Maximization Based Principal Component Analysis (EM-PCA), Singular Value Decomposition (SVD) and Power Spectral Density (PSD). After reducing the dimensions, the new obtained dimensionally reduced values are classified to get the epilepsy risk level from EEG signals with the help of a probabilistic model called Gaussian Mixture Model (GMM). The result analysis is performed with the benchmark terms like Performance Index, Accuracy, Quality Value and Time Delay. The most promising result in this study shows that when PSD is implemented as a dimensionality reduction technique and when classified with GMM, an average high accuracy of 97.46% is attained along with an average Performance Index of 94.69%.