An Explainable Transformer‐Based Model for Predicting Chronic Diseases Risk

ABSTRACT Chronic diseases are typically noncommunicable diseases and are the number one cause of death for human beings. With the continuous growth of medical data, how to effectively predict the risk of chronic diseases, particularly in the context of high‐dimensional medical data, has become an urgent issue that needs to be addressed. In this paper, we focus on three chronic diseases: fatty liver disease, diabetes disease and heart disease propose a Transformer‐based risk prediction model specifically designed for medical disease prediction—Med‐transformer. By applying interaction and attention mechanisms among different features, the model can automatically identify high‐risk factors for chronic diseases and provide reasonable explanations for the predictive results. The results demonstrate that the proposed model performs exceptionally well in predicting the risk of chronic disease. The model achieved an AUC of 0.9182 for predicting fatty liver disease, outperforming the XGBoost model (AUC = 0.9034). For heart disease prediction, our model achieved an AUC of 0.8357, outperforming the FT‐transformer model (AUC = 0.8222). For diabetes prediction, the model's AUC reached 0.9562, surpassing the RF model (AUC = 0.9500). Through the visualization of attention weights, we intuitively demonstrate the contribution of features to disease prediction, providing the medical field with more transparent and reliable decision‐making support. This model not only provides accurate predictions, but also generates a feature importance map that assists clinicians in identifying high‐risk individuals and understanding their risk composition, providing a new tool for achieving transparent and trustworthy AI‐assisted diagnosis.