A Case-Based Reasoning Method with Relative Entropy and TOPSIS Integration

Abstract

This paper addresses a new method of case-based reasoning (CBR). The aim of this work presented here is to provide effective warning knowledge for decision-makers. At first we design the similarity calculation methods according to the different case feature such as crisp number, interval number, crisp symbols and fuzzy linguistic variables. The similarity of each feature is calculated between target case and each historical case which step gets a similarity matrix. Then the CBR system employs a new ensemble measure for similarity matrix with two methods including relative entropy and the technique for order preference by similarity to an ideal solution (TOPSIS). On the basis, a new algorithm is designed, which is named as RTCBR. At the same time, RTCBR is tested on UCI data sets and compared with other two well-known CBR algorithms such as Euclidean distance CBR (ECBR) and Manhuttan distance CBR (MCBR). Empirical results indicate that RTCBR outperforms ECBR, MCBR, which can effectively improve the accuracy of CBR system.