From July 2009 to June 2011 I worked on mental workload measurement from brain signal analysis (EEG). The objectives was to develop a mental workload measurement tool that could be used in various applications. A possible scenario for application would be to monitor the mental workload of the members of a team of air traffic controllers, and use this information to decide to whom a new plane should be assigned to (e.g. the one with the lowest workload).
We collected EEG data from subjects who were performing a mental arithmetic task with 6 levels of difficulty, as well as in resting state (eyes opened and eyes closed). Spectral power was computed for each channel. Features were classified using a quadratic discriminant classifier and results were evaluated using a randomized 10-folds cross-validation procedure. Classification was conducted on a per subject basis, but results were averaged across subjects.
A workload index was constructed as follow. The output of the classifier was mapped to a workload value: 0 for the relaxed class, 2.5 for the low workload class, and 5 for the high workload class. These discrete values were then low passed filtered with a moving average filter over the past ten epochs, in order to obtain a smooth and continuous index of workload, ranging from 0 to 5. We measured the correlation between the continuous workload index, and the task difficulty, which was taken as the difficulty level for the task condition and as 0 for the other conditions.
We first evaluated classification accuracy for each pair of labels; mean accuracy was 96%. For all labels together (9 labels) the mean accuracy was 78.6%. The correlation coefficient between the workload index and the task difficulty was 0.94.
Brice Rebsamen, Trevor B. Penney, Kenneth Kwok, "EEG-based measure of cognitive workload during a mental arithmetic task", Communications in Computer and Information Science, 174 CCIS (PART 2) pp. 304-307, 2011.
Brice Rebsamen, Kenneth Kwok, Trevor B. Penney, "Evaluation Of Cognitive Workload From EEG A Mental Arithmetic Task", to be published in the proceedings of the 55th Annual Meeting of the Human Factors and Ergonomics Society, September 19-23, 2011.