Classification of Mental Workload Levels by Using EEG Signals

Mental workload is amount of the required cognitive capacity during performing tasks. Electroencephalogram (EEG) is an objective monitoring technique used to evaluate mental workload. In this study, feature extraction methods based on Katz's fractal dimension (KFD) and Higuchi's fractal dimension (HFD); and error correcting output coding (ECOC) are proposed to classify mental workload levels through EEG signals, which were recorded during performing of the simultaneous tasks. ECOC, which is a classifier combination technique proposed for multiclass classification problems, is employed to classify mental workload as low, moderate and high level. ECOC was created based on one vs. all approach, by using support vector machines (SVM), k nearest neighbourhood and quadratic discriminant analysis. The performance of the proposed method is evaluated on Simultaneous Task EEG Workload (STEW) dataset collected from 48 subjects. By using KFD and HFD with respectively, the classification accuracy was determined as %78.44 and %95.39; and Cohen's Kappa value was determined as 0.52 ve 0.89. The results indicate that combination of HFD and SVM-ECOC is a successful method in the multiclass classification of mental workload.