Data Fusion for Difficulty Adjustment in an Adaptive Virtual Reality Game System for Autism Intervention

A virtual reality driving simulator is designed as a tool for improving driving skills of individuals with autism spectrum disorders (ASD). Training at an appropriate driving difficulty level can maximize long term performance. Affective state information has been used for difficulty level adjustment in our previous work. This paper integrates performance with affective state information to predict the optimal difficulty level. The participant’s performance data, physiology signals, and eye gaze data are captured. The performance features and affective state features are extracted. Two classification methods, Support Vector Machine (SVM) and Artificial Neural Network (ANN), are implemented to predict difficulty level. The results demonstrate that performance together with affective state information outperform the separated features in difficulty level prediction. A highest accuracy of 83.09% is achieved with the integrated features.