Expert system for diagnosing diabetes disease using the dempster-shafer method based on patient symptoms

Abstract

Diabetes is a chronic disease that can cause serious complications and increase mortality risk if not detected and treated early. Therefore, an accurate and systematic diagnostic approach is needed to support early detection of diabetes. This study aims to implement the Dempster-Shafer method for classifying diabetes based on patients’ symptoms and clinical conditions. The dataset used in this research was obtained from Kaggle and consists of 768 patient records containing several clinical attributes related to diabetes diagnosis. The original numerical attributes were transformed into binary symptom representations to simplify the reasoning process in the expert system. Each symptom was assigned a confidence value or mass function representing the level of belief in the presence of diabetes. Furthermore, these confidence values were combined using the Dempster-Shafer combination rule to produce a final diagnostic decision while handling uncertainty in the available evidence.

The performance of the proposed method was evaluated using several classification metrics, including accuracy, precision, recall, and F1-score. The experimental results show that the Dempster-Shafer method achieved an accuracy of 68.23%, a precision of 53.31%, a recall of 72.01%, and an F1-score of 61.27%. The relatively high recall value indicates that the method is effective in identifying patients who are potentially affected by diabetes, which is important in early diagnosis applications. Based on these findings, the Dempster-Shafer method can be considered a feasible alternative approach for expert systems in supporting diabetes diagnosis and assisting medical decision-making in a more systematic and uncertainty-aware manner.