DET-VRET: a dataset for emotion detection in virtual reality using eye-tracking and machine learning

Purpose This research proposes an approach to classify emotions into four classes using eye-tracking data alone with machine learning in virtual reality (VR) stimuli with own VR datasets. Design/methodology/approach Firstly, we create an immersive VR database that is open to the public and can help investigations that use VR stimuli to identify emotions. Next, we use an eye-tracker to record and collect eye-tracking data. The emotions of a subject are stimulated by presenting an immersive 360-degree VR video using a VR headset. Finally, the performance of emotion detection is evaluated using three different machine learning classifiers, such as support vector machine, k-nearest neighbor and random forest. The results are compared from both intra-subject and inter-subject classifications with parameter tuning to obtain the best performance of emotion detection. Findings The fixation position has the best performance of emotion detection for both intra-subject and inter-subject classifications, which achieves the highest accuracy rate of 85.02% and 50.05%, respectively. Originality/value This study presents a novel multi-class classification of emotions using eye-tracking data with promising accuracy.