EEG signals classification based on mathematical selection and cosine similarity

Abstract EN

This paper presents a new electroencephalogram (EEG) signal classification using a fractal-cosine similarity approach for diagnosing epilepsy patients.

The proposed system provides two designed models with PSO as an optimization technique and without optimization.

A full classification design is achieved, including prepressing data by normalization, Particle Swarm Optimization (PSO) as optimization technique to reduce the features of EEG signals, Fractal metric computations, metric mapping, and cosine similarity for the final decision.

This paper used the BONN university EEG dataset, which consists of five categories.

The dataset was divided into four groups based on training set size and testing set size.

First, we are used to the training and testing ratio of 90/10.

The second case is 80/20, the third case is 70/30, and the final case is 60/40 respectively.

The proposed model achieves high rates of accuracy up to 100%.