Detection of Epileptic Seizure Based on ReliefF Algorithm and Multi-support Vector Machine

Abstract

In recent decades, epileptic seizure classification is the most challenging aspect in the field of health monitoring systems. So, a new system was developed in this research study for improving the accuracy of epileptic seizure classification. Here, epileptic seizure classification was done by using Bonn University Electroencephalogram (EEG) dataset and Bern-Barcelona EEG dataset. After signal collection, a combination of decomposition and transformation techniques (Hilbert Vibration Decomposition (HVD) and Dual-Tree Complex Wavelet Transform (DTCWT) was utilized for determining the subtle changes in frequency. Then, semantic feature extraction (permutation entropy, spectral entropy, Tsallis entropy, and hjorth parameters (mobility and complexity) were utilized to extract the features from collected signals. After feature extraction, reliefF algorithm was used for eliminating the irrelevant feature vectors or selecting the optimal feature subsets. A Multi-binary classifier: Multi-Support Vector Machine (M-SVM) was helpful in classifying the EEG signals such as ictal, normal, interictal, non-focal, and focal. This research work includes several benefits; assists physicians during surgery, earlier detection of epileptic seizure diseases, and cost-efficient related to the existing systems. The experimental outcome showed that the proposed system effectively distinguishes the EEG classes by means of Negative Predictive Value (NPV), Positive Predictive Value (PPV), f-score and accuracy.