An improved case-based reasoning method based on fuzzy clustering and mutual information

Case retrieval is the most critical link that affects the results of case based reasoning (CBR). Weights determination and attributes reduction are two key factors for case retrieval. They are studied separately and the relationship between them is ignored, which leads to the mismatch and finite precision issues. In order to solve this problem, it introduces an improved CBR method based on fuzzy clustering, mutual information and iterative learning strategy. Subtractive clustering and fuzzy c-means clustering are combined to divide case base into subspaces where case retrieval is conducted. Mutual information is used to evaluate the contribution of condition attributes to solutions, and iterative learning strategy is designed to update weights and realize attributes reduction at the same time. This hybrid method aims to improve the accuracy and efficiency of CBR. Simulation experiments based on UCI datasets and data from actual production of basic oxygen furnace are adapted to verify effectiveness of the proposed method.