Stacked Autoencoders Based Deep Learning Approach for Automatic Epileptic Seizure Detection

Abstract

Epilepsy is one of the major chronic nervous disorders, which affects the lives of millions of patients per annum globally, because of occurrence of sudden death or major injuries occurred during walk, driving or working in hazardous work environment. Its prognosis through modern technologies is the need of the day, which is attaining worldwide attention in research community with the use of latest technologies like internet of things, machine learning and cloud computing. This paper presents a model of automatic epileptic seizure detection model using Stacked Autoencoders based deep learning approach, which is an advanced form of machine leaning, employed for effectively handling the problem of big data with reduced complexity and processing time and to make this process more real time compatible with least delays. This model processes the sensed EEG signals by breaking it into short duration segments. Then, these EEG segments are fed to Stacked Autoencoders for its classification into different epileptic seizure stages like normal, preictal and ictal. The performance of this model has been compared with other existing models consisting of higher order spectral analysis based feature extraction and classification using traditional machine learning algorithms like Bayes Net, Naïve Bayes, Multilayer Perceptron, Radial basis function neural networks and C4.5 decision tree classifier. The analysis of performance through simulation results reveal that Stacked Autoencoders based deep learning approach is an efficient model for real time automatic epileptic seizures detection at early stage with classification accuracy 88.8%, sensitivity 89.44%, specificity 93.77% values and least value of processing time, which is approximately 23 times lesser than that of models utilizing traditional higher order statistics feature extraction and machine learning based classification approaches.