Section: New Results

Variable selection by genetic algorithm for the study of alertness states.

Participants : Marie Chavent, Laurent Vézard.

The aim of this work is to predict the state of alertness of an individual (binary variable, "normal" or "relaxed") from the study of brain activity (electroencephalographic signals EEG) collected with a limited number of electrodes. In fact, the set up of electrodes during the EEG signal acquisition is time consuming and these electrodes are correlated. In our study, the EEG of 58 participants in the two alertness states (116 records) were collected via a cap with 58 electrodes. After a data validation step based on the study of the contingent negative variation (CNV), 19 subjects were retained in the study. A CSP (Common Spacial Pattern) coupled to a linear discriminant analysis were used to build a decision rule and thus predict the alertness of the participants. A genetic algorithm was used to determine a subset of electrodes of size p '(where p' <p, where p = 58 is the number of electrodes). This presentation will present the CSP in the general framework and will introduce innovations made ​​to this method. The genetic algorithm will be described proposed and recent results will be presented.

These results have been obtained in collaboration with Pierrick Legrand of ALEA Inria team.

It has been presented in the Journée Évolutionnaire Thématique, 23éme édition [48] .