Optimizing Virtual Reality Eye Tracking Fixation Algorithm Thresholds Based on Shopper Behavior and Age

Abstract

Eye tracking (ET) is becoming a popular tool to study the consumer behavior. One of the significant problems that arise with ET integrated into 3D virtual reality is defining fixations and saccades, which are essential part of feature extraction in ET analysis and have a critical impact on higher level analysis. In this study, the ET data from 60 subjects, were recorded. To define the fixations, Dispersion Threshold Identification algorithm was used which requires to define several thresholds. Since there are multiple thresholds and extracted features, a Multi-Objective Reinforcement Learning (MORL) algorithm was implemented to solve this problem. The objective of the study was to optimize these thresholds in order to improve accuracy of classification of the age based on different visual patterns undertaken by the subject during shopping in a virtual store. Regarding the nature of the classification, the objective for this optimization problem was to maximize the differences between the averages of each feature in different classes and minimize the variances of the same feature within each class. For the current study, thresholds optimization has shown an improvement in results for the accuracies of classification between age groups after applying the MORL algorithm. In conclusion, the results suggest that the optimization of thresholds is an important factor to improve feature extraction methods and in turn improve the overall results of an ET study involving consumer behavior inside virtual reality. This method can be used to optimize thresholds in similar studies to provide improved accuracy of classification results.