Feasibility Study to Identify Machine Learning Predictors for a Virtual School Environment: Virtual Reality Stroop Task

An adaptive virtual school environment can offer cognitive assessments (e.g., Virtual Classroom Stroop Task) with user-specific distraction levels that mimic the conditions found in a student's actual classroom. Former iterations of the virtual reality classroom Stroop tasks did not adapt to user performance in the face of distractors. While advances in virtual reality-based assessments provide potential for increasing assessment of cognitive processes, less has been done to develop these simulations into personalized virtual environments for improved assessment. An adaptive virtual school environment offers the potential for dynamically adapting the difficulty level (e.g., level and amount of distractors) specific to the user's performance. This study aimed to identify machine learning predictors that could be utilized for cognitive performance classifiers, from participants (N = 60) using three classification techniques: Support Vector Machines (SVM), Naive Bayes (NB), and k-Nearest Neighbors (kNN). Participants were categorized into either high performing or low performing categories based upon their average calculated throughput performance on tasks assessing their attentional processes during a distraction condition. The predictors for the classifiers used the average cognitive response time and average motor response dwell time (amount of time response button was pressed) for each section of the virtual reality-based Stroop task totaling 24 predictors. Using 10-fold cross validation during the training of the classifiers, revealed that the SVM (86.7%) classifier was the most robust classifier followed by Naive Bayes (81.7%) and KNN (76.7%) for identifying cognitive performance. Results from the classifiers suggests that we can use average response time and dwell time as predictors to adapt the social cues and distractors in the environment to the appropriate difficulty level for the user.