Towards a Selection Mechanism of Relevant Features for Automatic Epileptic Seizures Detection

Epilepsy diagnosis is frequently confirmed using electroencephalogram (EEG) along with clinical data. The main difficulty in the diagnosis is associated with the large amount of data generated by EEG, which must be analyzed by neurologists for identifying abnormalities. One of the main research challenges in this area is the identification of relevant EEG features that allow automatic detection of epileptic seizures, especially when a large number of EEG features are analyzed.

The aim of this paper is to analize the accuracy of algorithms typically used in feature selection processes, in order to propose a mechanism to identify a set of relevant features to support automatic epileptic seizures detection.

This paper presents a set of 161 features extracted from EEG signals and the relevance analysis of these features in order to identify a reduced set for efficiently classifying EEG signals in two categories: normal o epileptic seizure (abnormal). A public EEG database was used to assess the relevance of the selected features. The results show that the number of features used for classification were reduced by 97.51%.

The paper provided an analysis of the accuracy of three algorithms, typically used in feature selection processes, in the selection of a set of relevant features to support the automatic epileptic seizures detection. The Forward Selection algorithm (FSA) produced the best results in the classification process, with an accuracy of 80.77%.