Improved Epilepsy Detection method by addressing Class Imbalance Problem

Abstract

Early and reliable detection of neurological disorders is important for effective treatment of patients. In spite of reasonable amount of research done in the field of early detection of epileptic seizure, still an effective model for predicting the same is absent. Motivated by this, in the current study the class imbalance problem associated with classification of patients into healthy and epilepsy affected ones is addressed. Two well established algorithms namely Synthetic Minority Oversampling Technique (SMOTE) and Selective Pre-Processing of Imbalanced Data Algorithm (SPIDER) have been used in order to combat the imbalanced classes. Afterwards, three different classifiers namely KNN, SVM and MLP-FFN have been used for the classification task. Experimental results revealed that addressing imbalances classes improved the classification accuracy to a greater extent.