Towards Automatic Detection of Social Anxiety Disorder via Gaze Interaction

Abstract

Social anxiety disorder (SAD) is an extreme fear of underperformance in various social situations. It is necessary to detect people with or without SAD for counseling and treatment. A few manual techniques in the existing literature show the possibility of SAD detection from gaze interaction. However, an automated prediction of SAD is scarce. In this research, an automatic technique to predict SAD using gaze interaction/avoidance is proposed, where a custom application was developed that used the Haar Cascade classifier to predict gaze interaction/avoidance. The experiments were conducted on 50 participants in a live environment using the developed application. SAD classes were predicted by using decision tree classifiers from the created gaze dataset. The results proved that SAD could be predicated with an overall accuracy of 80%. Furthermore, four classes of SAD (Mark, Moderate, Severe, Very Severe along with ‘No SAD’) could be predicted with an accuracy of 80%, 70%, 90%, 80%, and 80%, respectively. The research proved the possibility to predict SAD using computer-based methods without human intervention. Furthermore, it created the possibility of aiding a subjective Liebowitz Social Anxiety Scale (LSAS) with an objective technique described in this research.