Classification of Epileptic and Non-Epileptic Electroencephalogram (EEG) Signals Using Fractal Analysis and Support Vector Regression

Abstract

Seizures are a common symptom of this neurological condition, which is caused by the discharge of brain nerve cells at an excessively fast rate. Chaos, nonlinearity, and other nonlinearities are common features of scalp and intracranial Electroencephalogram (EEG) data recorded in clinics. EEG signals that aren't immediately evident are challenging to categories because of their complexity. The Gradient Boost Decision Tree (GBDT) classifier was used to classify the majority of the EEG signal segments automatically. According to this study, the Hurst exponent, in combination with AFA, is an efficient way to identify epileptic signals. As with any fractal analysis approach, there are problems and factors to keep in mind, such as identifying whether or not linear scaling areas are present. These signals were classified as either epileptic or non-epileptic by using a combination of GBDT and a Support Vector Regression (SVR). The combined method's identification accuracy was 98.23%. This study sheds light on the effectiveness of AFA feature extraction and GBDT classifiers in EEG classification. The findings can be utilized to develop theoretical guidance for the clinical identification and prediction of epileptic EEG signals. Doi: 10.28991/ESJ-2022-06-01-011 Full Text: PDF

Keywords

• hurst exponent
• adaptive fractal analysis
• eeg
• support vector regression
• gradient boosting decision trees.