Highly explainable cumulative belief rule-based system with effective rule-base modeling and inference scheme

Advancement and application of rule-based expert systems have been a key research area in explainable artificial intelligence (XAI) because the rule-base is one of the most common and natural explainable frameworks for knowledge representation. The present work aims to design a novel rule-based system, called Cumulative Belief Rule-Based System (CBRBS), by establishing efficient rule-base modeling and inference procedures, where the rule-base modeling procedure includes the generation of cumulative belief rules via numeric data transformation and extended belief rule integration, and the rule-base inference procedure includes the inference of cumulative belief rules via consistent rule activation and activated rule integration. All these procedures enable CBRBS to better achieve the balance of explainability, high-efficiency (or computing complexity), and accuracy to fit with different application scenarios, as well as overcome the limitations of classical rule-based systems. Extensive experiments based on the well-known pipeline leak detection problem and open-source classification problems are conducted to illustrate the feature and advantage of the CBRBS over other classical rule-based systems and some commonly used classifiers.