Seven Epileptic Seizure Type Classification in Pre-Ictal, Ictal and Inter-Ictal Stages using Machine Learning Techniques

Abstract

Background: Epileptic Seizure type diagnosis is done by clinician based on the symptoms during the episode and the Electroencephalograph (EEG) recording taken during inter-ictal period. But main challenge is, most of the time with the absence of any attendee, the patients are unable to explain the symptoms and not possible to find signature in inter-ictal EEG signal. Aims: This paper aims to analyze epileptic seizure Electro-encephalograph (EEG) signals to diagnose seizure in pre-ictal, ictal and inter-ictal stages and to classify into seven different classes. Methods: Temple University Hospital licensed dataset is used for study. From the seizure corpus, seven seizure types are pre- processed and segregated into pre-ictal, ictal and inter-ictal stages. The multi class classification performed using different machine and deep learning techniques such as K- Nearest Neighbor (KNN) and Random Forest, etc. Results: Multiclass classification of seven type of epileptic seizure with 20 channels, with 80-20 train-test ratio, is achieved 94.7%, 94.7%, 69.0% training accuracy and 94.46%, 94.46% 71.11% test accuracy by weighted KNN for pre-ictal, ictal and inter-ictal stages respectively. Conclusion: Seven epileptic seizure type classification using machine learning techniques carried out with MATLAB software and weighted KNN shows better accuracy comparatively.