Development of A Deep Learning Model for Mental Health Classification and Early Screening through Draw a Person (DAP) Test Images

Abstract

Mental health, as defined by the World Health Organization (WHO), is a fundamental aspect of overall well-being. The increasing complexity of modern society, coupled with rising levels of competition and stress, significantly impacts individuals’ mental health. The DAP test is a psychological assessment tool that uses human figure drawings to gain insights into an individual’s personality and mental condition. YOLO (You Only Look Once) is a deep learning algorithm based on Convolutional Neural Networks (CNNs) designed for real-time object detection. This study utilizes a DAP image dataset contributed by adolescents aged 12 to 16 years to develop a model for detecting and classifying objects in DAP images using the YOLOv8 algorithm. Optimal training results were achieved after 150 epochs, yielding a Precision of 0.821, Recall of 0.799, and mAP50 of 0.88. The model evaluation demonstrated an F1-Score of 0.78, indicating a balanced performance between Precision and Recall. Psychological analysis was conducted based on symptoms extracted from the characteristics of DAP images. Mental health conditions were classified according to severity levels consisting of minor, medium, and serious, based on weighted symptomatology derived from DAP image characteristics. The successful development of this model highlights its capability to classify various mental health conditions based on psychological analysis of DAP images. The findings suggest that mental health classification using DAP test images has the potential to support early screening and psychological assessment by providing an innovative and objective approach to identifying psychological indicators.