Design and Development of a Mental Disorder Diagnosis Expert System Using the Dempster-Shafer Method

Abstract

An information system designed to provide comprehensive information regarding employee transfers and retirement Mental disorders represent a major global health challenge with continuously increasing prevalence. According to the World Health Organization (WHO) 2022 data, approximately one billion people worldwide live with mental disorders, yet access to psychiatric services remains severely limited, particularly in developing countries such as Indonesia. The psychiatrist-to-population ratio in Indonesia is less than one per 100,000 people, far below the WHO minimum standard of three psychiatrists per 100,000 population. This situation highlights the urgent need for artificial intelligence-based technology that can assist in the early screening and diagnosis of mental disorders independently and efficiently. This study aims to design and develop an expert system for diagnosing mental disorders using the Dempster-Shafer method. The Dempster-Shafer method was selected for its capability to handle the inherent uncertainty in clinical diagnosis processes, where a patient may exhibit overlapping symptoms across different disorders. Unlike classical probabilistic approaches, Dempster-Shafer does not require complete prior probabilities, making it more suitable for imperfect clinical data. The knowledge base was constructed through in-depth interviews with three experienced psychiatrists and referenced against DSM-5 criteria. The system covers seven types of mental disorders—Major Depression, Generalized Anxiety Disorder, Bipolar Disorder, Schizophrenia, Obsessive-Compulsive Disorder (OCD), Post-Traumatic Stress Disorder (PTSD), and Panic Disorder—with a total of 51 defined symptoms. The system was developed as a web-based application using the PHP Laravel framework and MySQL. Evaluation was conducted using 50 test cases confirmed by psychiatrists, yielding an average accuracy of 84.29%, with precision, recall, and F1-Score consistently above 80% for all disorder types. The results demonstrate that the Dempster-Shafer method is effective in handling symptom ambiguity and provides representative belief values for diagnostic recommendations.