Integrating Bayesian Theory and Fuzzy Logics with Case-Based Reasoning for Car-Diagnosing Problems

Abstract

This paper integrates fuzzy Bayesian classifier with case-based reasoning (CBR) for improving case retrieval performance. In traditional retrieval process, CBR uses similarity metric to calculate similarity degree between all former cases and new case. Then, it retrieves the highest one. However, it would cause computation complexity and waste time. Our proposed approach uses Bayesian classifier to classify case base, then retrieve similar cases. In the real world problems, attributes may not always be discrete; they might be continuous as well. When users encounter continuous attributes, they must define an appropriate conditional likelihood density function. In fact, defining an appropriate conditional likelihood function is difficult. The issue of defining conditional likelihood density function of continuous or attributes is critical. Besides, the choice of likelihood density function would affect the accuracy of Bayesian network. Therefore, we apply fuzzy set theory to Bayesian network to define an appropriate conditional likelihood density function. An efficient fuzzy Bayesian case-based reasoning (FB-CBR) was built to solve car-diagnosing problems. The result shows that our proposed model is better than conventional models both in case matching accuracy and computational efficiency.