An Innovative Information-Based Strategy for Epileptic EEG Classification

Epilepsy is one of the most prevalent neurological diseases. Electroencephalography (EEG) is an essential tool for diagnosing and detecting epilepsy. In this paper, we aimed to provide an accurate system for classifying epileptic EEG signals using an information fusion-based approach. To this effect, an innovative feature-level fusion methodology was presented. First, utilizing the theory of information, the EEG signals were characterized by two similarity indicators, including correntropy and Cauchy-Schwarz divergence. Then, different combining modes, including, stacking, adding, weighted sum, and maximum rules were used to combine the information of both similarity measures. Concurrently, a novel algorithm was also proposed to define the weights of each feature based on the information gain ratio (IGR). Finally, the probabilistic neural network and the k-nearest neighbor were implemented to detect the epileptic conditions. The EEGs of the Bonn database were used to appraise the framework. Promising classification rates were obtained for epilepsy detection. The results revealed the potent ability of the fusion approach based on maximum IGR in the improvement of the classifier performances.