An approach to detect epileptic seizure using XAI and machine learning

Abstract

One of the most common neurological disorder in health sector is Epileptic Seizure (ES) which is occurred by sudden repeated seizures. Hitherto more than 50 million people in the whole world are suffering from Epileptic Seizures. The abnormal brain activity of the central nervous system often causes unusual behavior, losing awareness and psychological problems etc. Moreover, many risks associated with epileptic seizures include sudden unexpected death in epilepsy (SUDEP) which is really a concerning problem discussed in this article. For abstaining from adverse consequences of epileptic seizure-like this health sector focuses more on the early prediction and detection of epilepsy. The complex signals of brain activity are reflected as swift-passing exalted peaks in Electroencephalogram (EEG). Initially, the specialist inspects the EEG signals over a few weeks or months to identify the presence of epileptic seizures, which is a very time-consuming and challenging task. Hence, Machine learning (ML) based classifiers are capable to categorize EEG signals and detect seizures along with displaying related perceptible patterns by maintaining accuracy and efficiency. In order to detect epileptic seizures, EEGbased signal recognition algorithms had been shown in this paper by applying both Multi-Class Classification and Binary classification. The algorithms were Decision Tree Algorithm, Random Forest Algorithm, Multi-Layer Perceptron (MLP) and K-Nearest Neighbor (KNN), Gradient Boosting Classifier, Gaussian Na¨ıve Bayes, Complement Na¨ıve Bayes, SGD Classifier, Explainable Artificial Intelligence (XAI), LIME Algorithm etc. However, K-Nearest Neighbor appears with pretty higher accuracy in certain conditions.