Machine learning classification analysis for an adaptive virtual reality Stroop task

Advances in virtual environment (VE) technologies have afforded psychologists with high-dimensional virtual reality (VR) platforms that enhance the complexity and dimensionality of cognitive assessments. The Virtual Reality Stroop Task HMMWV (VRST; Stroop stimuli embedded within a virtual high mobility multipurpose wheeled vehicle) is a VR assessment involving both cognitive and affective components. There is a need for adaptive virtual environments (AVEs) that can adjust the complexity of environmental stimuli relative to the way the participant is performing. To develop the VRST into an AVE assessment, classifier algorithms must be developed. While previous research has explored classifier algorithms for modeling arousal and cognitive performance in the VRST, machine learning (ML) classifiers have not been developed for an adaptive VRST. The current study developed ML classifiers for an adaptive version of the VRST. The assessment of Naive Bayes (NB), k-Nearest Neighbors (kNN), and Support Vector Machines (SVM) machine learning classifiers found that SVM and NB classifiers tended to have the highest accuracies and greatest areas under the curve when classifying users as high or low performers. The kNN algorithms did not perform as well. As such, SVM and NB may be the best candidates for creation of an adaptive version of the VRST.