Determination of Increased Mental Workload Condition From EEG by the Use of Classification Techniques

Abstract

Extraction of the information hidden in the brain electrical signal enhance the classification of the current mental status. In this study, 16 channel EEG data were collected from 15 volunteers under three conditions. Participants were asked to rest with eyes open and eyes closed states each with a duration of three minutes. Finally, a task has been imposed to increase mental workload. EEG data were epoched with a duration of one second and power spectrum was computed for each time window. The power spectral features of all channels in traditional bands were calculated for all subjects and the results were concatanated to form the input data to be used in classification. Decision tree, K-nearest neighbor and Support Vector Machine techniques were implemented in order to classify the one second epochs. The accuracy value obtained from KNN was found to be 0.94 while it was 0.88 for decision tree and SVM. KNN was found to outperform the two methods when all channel and power spectral features were used. In can be concluded that, even with the use of input features formed by concatanating all subject’s data, high classification accuracies can be obtained in the determination of the increased mental workload state.