Optimized Feature Selection for Enhanced Epileptic Seizure Detection

Abstract

Genetic algorithm (GA) based feature selection method is an evolving search heuristic, used to provide solutions to optimization problems. Feature selection is an important aspect that improves classification accuracy. The main objective of this work is to utilize GA for feature selection by integrating it with a bank of multi-class Support Vector Machine (SVM) for identification of the effective feature set. The proposed GA based approach finds its application in epileptic seizure detection. EEG dataset containing artefacts and noise were removed by employing constrained Independent Component Analysis (cICA) and Stationary Wavelet Transform (SWT). The features of the input data are constructed in the form of feature vector by FastICA technique. The fitness calculation for the selection of individuals in the GA is calculated by a Linear Discriminant Analysis (LDA) classifier. The multi-class Support Vector Machine (SVM) (one-against-all) classifier is used for the validation of the selected features. The samples are taken from 948 patients and the classes are divided as normal, seizure, and seizure-free using artificial neural networks. Experimental results shows that the GA - multi-SVM feature selection technique can achieve higher accuracies as compared to the case without feature selection. Keywords: EEG (Electroencephalogram), Electrocardiogram (EKG), Electromyogram (EMG), EOG (Electro-oculogram), Genetic Algorithm (GA), Linear Discriminant Analysis (LDA) and Support Vector Machines (SVM).