Classification of Epileptic Seizures using Recurrence Plots and Machine Learning Techniques

Abstract

Epilepsy is a neurological disorder noticeable by sudden repeated episodes of sensory commotion related with anomalous electrical activity in the brain. Electroencephalogram (EEG) based seizure detection system is a promising non-invasive diagnosis option for refractive epilepsy. EEG signals in this system are decomposed and classified to improve the diagnostic accuracy as compared to human diagnosis. There is a need to explore reliable and faster classification algorithms to develop real time seizure detection system. Due to non linear nature of EEG, non-linear techniques are used to examine the EEG signals. The paper proposes a non linear technique of EEG feature extraction based on the Recurrence Plots (RP), and Recurrence Quantification Analysis (RQA) parameters derived from the RP have been used to categorize the EEG signal information as pre-ictal, ictal and normal classes. RP is an advanced technique of nonlinear data analysis and RQA parameters of RP compute the significant features of signals. These features have been classified using Artificial Neural Network (ANN), Probabilistic Neural Network (PNN) and Support Vector Machine (SVM). In order to improve the current information transfer rate of seizure detection system, the SVM which classifies EEG signals with highest speed with accuracy of 91.2 % is selected for classification.