A Comparison of Performance and Psychophysiological Classification of Complex Task Performance

Abstract

Psychophysiologically guided classifiers have been used to distinguish between low and high mental workload conditions. Prior studies used cognitive tasks that were quite different in terms of the cognitive demands placed on the operator. It is not clear if these classifiers can discriminate between tasks that are more similar in cognitive demand. A complex uninhabited aerial vehicle simulator provided conditions that were similar but placed emphasis on different aspects of the task. Performance and subjective workload data could only discriminate between the baseline task and the three more difficult tasks. Artificial neural networks (ANN) that used psychophysiological data were able to discriminate among the four tasks with a mean accuracy of 83.4%. ANNs that used a larger number of features and saliency analysis produced higher classification accuracies than ANNs using fewer features. It appears that when comparing cognitively similar complex tasks, given sufficient information, ANNs are capable of finer discrimination than performance and subjective workload measures.