[프로그램]WZT 그림분석을 딥러닝을 활용하여 청소년 폭력성 예측 프로그램

Abstract

In adolescence, changes in stress can lead to negative behavior in daily life, and it is important to control negative problem behavior through appropriate coping mechanisms. A self-reported questionnaire for adolescents has the disadvantage of being difficult to obtain necessary information due to defensive or insincere responses. Alternatively, a projective test using pictures can provide adolescents with the necessary information through direct experiences that the subject unconsciously reacts to and is represented through pictures. There are few methods for analyzing images drawn by adolescents as image data. In this study, we directly utilized image data expressed by adolescents through the Wartegg-Zeichentest (WZT) and analyzed it to predict adolescent violence. This study aims to analyze data from 134 violent students who have received fifth-degree punishment for violent behavior and 134 non-violent students. The subjects of this study are students who have received punishment for violence at special schools (punishment 5). We will use CNN, CNN(SVM), CNN(SVM)+Style transfer GAN, and Ensemble to analyze the violent drawing images of WZT and predict the results through deep learning. As a result, we can predict the violence of the pictures with an accuracy of 93-98%. This study is the first to analyze and predict the violence level of WZT automatically, drawing images directly drawn by teenagers using a deep learning model. The remarkable development of deep learning in extracting features from images is expected to create more opportunities for further research.