Classification of Disease with Explainable Artificial Intelligence

Abstract

With the advancement of technology in recent years, it has become possible to perform fast and accurate symptom-disease classification in medical diagnosis processes. However, although the accuracy of these models has increased, their decision-making processes are often incomprehensible due to their so-called 'black box' structures, raising questions about their reliability. In this study, disease-symptom classification was performed using DistilBERT and ClinicalBERT+BioBERT models, and the classification and decision-making processes were then analyzed using modern Explainable Artificial Intelligence methods such as SHAP, LIME, and Integrated Gradients. In this way, the decision mechanisms of the models are made more transparent, interpretable, and reliable, providing a significant contribution to the adoption and clinical use of medical artificial intelligence systems. © 2025 IEEE.