ANALYSIS OF GENETIC ALGORITHM DRIVEN AUTOENCODERS FOR EPILEPSY CLASSIFICATION USING CERTAIN POST CLASSIFIERS

The transient abnormal behaviour of the neurons represents the seizures in the context of epilepsy. The constant panic of the seizure occurrence and a feeling of restlessness and helplessness have a strong impact on the quality of life of epileptic patients. To improve the quality of life of epileptic patients who are drug resistant, a patient specific algorithm is mandatory to predict the seizures based on Electroencephalogram (EEG) signals with a very high specificity and sensitivity value before the seizure occurrence. Due to its excellent temporal resolution and low maintenance cost EEG is used to capture brain signals. The EEG can depict the state of a person such as awake state, sleep state or anaesthetized state etc. The characteristic patterns of the electrical potentials of one state differ from the characteristic patterns of the electrical potentials of other states. Thus EEG plays a vital role in field like electro-physiology assessment, spike wave detection, diagnosing epilepsy, detection of early brain tumour, monitoring the sleep analysis and in depth stage of anesthesia. As EEG signal is a stochastic complex non-stationary signal and recorded for a long time, it is hectic to reduce the dimensions and extract the important features in either time domain or frequency domain. In this paper, the Autoencoders with the application of Genetic Algorithm (GA) are used to reduce the dimensions of the EEG data set and then the risk of epilepsy is classified with the help of Post classifiers like Approximate Entropy Model (AEM), Gaussian Mixture Model (GMM) and Naïve Bayesian Classifier (NBC) Model. The Performance metrics are thoroughly analyzed in terms of accuracy, time delay values, performance index, specificity, sensitivity and quality values. Results show that an average accuracy analysis of 97.951% is obtained when it is classified with GMM.