An automated diagnosis of epilepsy using EEG derivation based on few features

Pre-processing and feature extraction are essential steps for EEG signal classification based on Machine Learning. An appropriate choice of signal processing methodology in these two steps can perfectly improve classifier performance. Different approaches for signal decomposition, transformation and feature extraction are used in the literature to extract useful and relevant information from EEG signals and remove redundant and irrelevant ones. In this paper, we propose an efficient method for EEG analysis. We demonstrate that the variances and correlation dimension calculated from the EEG signals and their derivatives (First, second and third derivatives) allow producing a small size features space, four features, with high relevance for epilepsy diagnosis using two public EEG databases, Bonn and New Delhi. The proposed approach exceeds state of the art methods' performances, high accuracy, sensitivity, and selectivity, were achieved using k-Nearest Neighbor KNN.