Practical Integrated Weighted Approximate Reasoning

Pattern-matching-based approximate reasoning methods (e.g. CRI) can be effective and efficient when applicable, while FRI methods work well when facing incomplete knowledge bases. To maximise the benefits of both approaches, this chapter introduces a combined technique to jointly exploit the advantages of CRI for matched observations and those of FRI for unmatched ones. Particularly, conventional CRI and transformation-based FRI are united within a single system framework. It showcases how to develop an approximate reasoning system that exploits an attribute-weighted rule base with enhanced CRI and FRI inference mechanisms. It illustrates that such an organically combined fuzzy rule-based system is capable of deriving more accurate and less time-consuming inference outcomes. In particular, it utilises an implemented computational mechanism, W-Infer-polation, to explain the inference processes concerned, depicting a general framework for building fuzzy systems by integrating CRI and FRI. The chapter also provides an analysis of the benefits of conducting inference rule selection with attribute weights, showing the efficacy of an integrated system in performing approximate reasoning for various practical problem cases.