A Chronic Kidney Disease Diagnostic Model Based on an Interpretable Deep Belief Rule Base

Chronic Kidney Disease (CKD) has become a serious public health problem because of its characteristic ’three highs and one low’: high prevalence, high disability rate, high medical costs, and low awareness. Therefore, an accurate diagnosis of CKD is crucial. Given the unique nature of the medical field, it is essential to ensure that the results of the CKD diagnostic model are trustworthy to doctors and patients and are practically applicable. Belief Rule Base (BRB) models explain their results through transparent reasoning processes and belief distributions. However, existing BRB models are typically constructed as one-time setups that require reconstruction to include new patient indicators. Furthermore, redundant rules within BRB models increase the difficulty of optimization, and randomness in the optimization process can affect the interpretability of the model. To address these challenges, this paper presents a CKD diagnostic model based on an interpretable Deep Belief Rule Base (DBRB-I). The model leverages a deep BRB structure to enhance scalability and prevent rule explosions. Additionally, it introduces a novel rule reduction method that incorporates expert knowledge to simplify redundant rules in an interpretable way, thus easing the optimization process. To preserve the model’s interpretability, the optimization algorithm includes constraints grounded in expert knowledge. Case studies and comparisons with seven other methods show that the CKD diagnostic model based on DBRB-I outperforms the others in a practical diagnostic setting.