A Fractal based Machine Learning Method for Automatic Detection of Epileptic Seizures using EEG

Abstract

Epilepsy is one of the most common and chronic brain disorders that causes recurrent seizures. In this paper, a machine learning method utilizing non-linear features and a Support vector machine (SVM) classifier is proposed for the automatic detection of epileptic seizures using electroencephalographic (EEG) signals. In this proposed model, Hurst exponent and logarithmic Higuchi fractal dimension (HFD) non-linear features are extracted from EEG signals which are then classified using SVM and K-nearest neighbor (KNN) classifiers. For the model’s implementation and performance evaluation, a publicly available CHB-MIT EEG dataset is used. The proposed model uses the Hurst component, logarithmic HFD, and SVM classifier, resulting in an average accuracy of 99.81%, Recall 100%, and TNR 0.99. Similarly, the proposed model utilizing the Hurst component, logarithmic HFT, and KNN classifier resulted in an accuracy of 93.21%, recall 92.56%, and Tnr 0.92. This automated and highly accurate model can be implemented in remote-based applications using the Internet of Medical Things (IoMT) framework.