Statistical Performance Effect of Feature Selection Techniques on Eye State Prediction Using EEG
Pages 224-230
Jean de Dieu Uwisengeyimana, Nusaibah Khalid Al_Salihy and Turgay Ibrikci
DOI: http://dx.doi.org/10.6000/1929-6029.2016.05.03.9
Published: 16 July 2016

Abstract: Several recent studies have demonstrated that electrical waves recorded by electroencephalogram (EEG) can be used to Predict eye state (Open or Closed) and all the studies in the literatures used 14 electrodes for data recording. To reduce the number of electrodes without affecting the statistical performance of an EEG device, it is not an easy task. Hence, the focus of this paper is on reducing the number of EEG electrodes by means of feature selection techniques without any consequences on the statistical performance measures of the earlier EEG devices. In this study, we compared different attribute evaluators and classifiers. The results of the experiments have shown that ReliefF attribute evaluator was the best to identify the two least important features (P7, P8) with 96.3% accuracy. The overall results show that two data-recording electrodes could be removed from the EEG devices and still perform well for eye state prediction. The accuracy achieved was equal to 96.3% with KStar (K*) classifier which was also the best classifier among the 21 tested classifiers in this study.