A. Gelbukh and C.A. Reyes-Garcia (Eds.): MICAI 2006, LNAI 4293, pp. 176 – 185, 2006. © Springer-Verlag Berlin Heidelberg 2006 A Connectionist Fuzzy Case-Based Reasoning Model Yanet Rodriguez 1 , Maria M. Garcia 1 , Bernard De Baets 2 , Carlos Morell 1 , and Rafael Bello 1 1 Universidad Central de Las Villas, Carretera a Camajuani km 51/2, Santa Clara, Cuba {yrsarabia, mmgarcia, cmorellp, rbellop}@uclv.edu.cu 2 Ghent University, Coupure links 653, B-9000 Gent, Belgium Bernard.DeBaets@UGent.be Abstract. This paper presents a new version of an existing hybrid model for the development of knowledge-based systems, where case-based reasoning is used as a problem solver. Numeric predictive attributes are modeled in terms of fuzzy sets to define neurons in an associative Artificial Neural Network (ANN). After the Fuzzy-ANN is trained, its weights and the membership degrees in the training examples are used to automatically generate a local distance function and an attribute weighting scheme. Using this distance function and following the Nearest Neighbor rule, a new hybrid Connectionist Fuzzy Case-Based Rea- soning model is defined. Experimental results show that the model proposed al- lows to develop knowledge-based systems with a higher accuracy than when using the original model. The model takes the advantages of the approaches used, providing a more natural framework to include expert knowledge by us- ing linguistic terms. 1 Introduction Case-Based Reasoning (CBR) may be defined as a model of reasoning that incorpo- rates problem-solving understanding and learning integrated with memory proc- esses. CBR can mean different things depending on the intended use of the reason- ing: adapt and combine old solutions to meet new demands or use old cases to explain new situations or to justify new solutions. CBR can be classified into two major types: problem solving CBR and interpretative CBR [1]. A model to build hybrid Knowledge-Based Systems (KBS), where CBR has the functionality to avoid the non-existing explanation facilities of the connectionist approach is presented in [2]. That model is a variant of the model of Stanfill and Waltz [3], in which an Artificial Neural Net (ANN) is used to suggest the value of the target attribute for a given query. The case-based module uses a similarity function to justify the solu- tion given by the ANN, which includes ANN weights. Hereafter, this model will be referred as original model. The original model uses a simple implementation of the Interactive Activation and Competition neural net model proposed by Rumelhart in [4], which is referred as SIAC. The attributes used to define cases can be both numeric and symbolic types. When a numeric attribute is used, many different values for it should appear in the case base. Therefore, the quantity of neurons in the ANN will increase very rapidly when a numeric attribute is used. In most cases, however, it would be enough to consider some values likely to represent a group of values close to them. Another way