Probe-Independent EEG Assessment of Mental Workload in Pilots
Abstract
We developed probe-independent algorithms for classifying three levels of task-complexity based on 4-channel electroencephalographic (EEG) recordings during simulated flight. Using a library of 168 input features drawn from different signal processing application domains, we evaluated 10 different classifiers, using 10-fold cross-validation to estimate generalization performance. The best subsets of features for each subject yielded a median classification accuracy of 92.81%, with 100% accuracy in two subjects and greater than 70% in all 19 subjects. Generally, the EEG line length and linear discriminant analysis were among the most effective features and classifiers, respectively. However, to maximize performance, the feature set-classifier combinations should be chosen based on the individual. No single channel proved more valuable than another in predicting flight task-complexity, but fusing the information across channels improved performance in 18 of 19 subjects. Given the success we had in producing high classification accuracies without an auditory stimulus, we believe this algorithm may be useful in developing optimal equipment or training techniques to minimize mental workload, and/or to monitor the mental state of a pilot over the course of a mission.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 18, 2015
- Accession Number
- ADA619147
Entities
People
- Michael K. Johnson
Organizations
- United States Naval Academy