Hierarchal Machine Learning Approach To Explore Automatic Seizure Detection In EEG

Abstract

Epilepsy is one of the neurological disorders which appears due to the recurrence of unexpected sudden reactions of the brain, also called as seizures. Electroencephalogram (EEG) is most commonly used test measure to track the ongoing electrical activity of brain and it is widely used in analysis and detection of electro epileptic seizures. It is often difficult to identify the brain subtle changes in EEG Datasets because of its complexity. Aicardi syndrome is one of ophthalmic findings which involves pigmentation of retinal lesions that found in Occipital lobe epileptic patients. Aicardi affected patients retinal image evidences retinal lesions. Based on this idea here we proposed Hierarchal Machine Learning Approach (HMLA) which is the combination of Grasshopper Optimization Algorithm (GOA) and Support vector machine (SVM) for automatic seizure detection based on the amount of lesions portion present in retinal images. It involves identifying the amount of lesion information in retinal images and thus classify such data as either epileptic data or nonepileptic data by using radial bias kernel function classifiers with different notations present in such images. To enable this Grasshopper Optimization Algorithm is used to explore effective subset of features and then optimal parameters based on SVM for successful classification of seizure in retinal images. Further improvement of proposed approach, it gives better seizure classification and enhance diagnosis of epilepsy with effective accuracy 90-100 % with comparison of non-epileptic retinal image data. Experiments of proposed approach gives better and efficient results when comparing to existing approaches