European and Mediterranean Conference on Information Systems (EMCIS) 2006, July 6-7 2006, Costa Blanca, Alicante, Spain Juan.L. Castro, María Navarro SIMILARITY LOCAL ADJUSTMENT: INTRODUCING ATTRIBUTE RISK INTO THE CASE SIMILARITY LOCAL ADJUSTMENT: INTRODUCING ATTRIBUTE RISK INTO THE CASE Juan L. Castro, Department of Computer Science and Artificial Intelligence, ETSI Informática ,Granada University, castro@decsai.ugr.es María Navarro, Department of Computer Science and Artificial Intelligence, ETSI Informática, Granada University, minj@correo.ugr.es José M. Sánchez, Department of Computer Science and Artificial Intelligence, ETSI Informática, Granada University, jmsa@decsai.ugr.es José M. Zurita, Department of Computer Science and Artificial Intelligence, ETSI Informática, Granada University, zurita@decsai.ugr.es Abstract The method used to evaluate the overall similarity between cases in the CBR retrieval stage plays a very important role when deciding the case to select and therefore the final solution to apply. Many case retrieval techniques have been developed and perhaps the most popular is the one which uses the nearest-neighbour (NN) matching function. This technique first calculates the similarity between the target problem and an old case regarding individual attributes, and then places the overall similarity of the target problem with the old case. The overall similarity is assessed by a weighted sum of all the similarity measures between the attributes, where the weighting factor represents the degree of importance of each attribute. Since this factor is sometimes allocated by a human expert and other times by the human user, the weight assessments usually involve human subjectivity. In order to prevent this subjectivity from affecting case retrieval and to avoid losing the valuable information obtained when measuring the risk in applying the solution of the old case to the current case, a new variable is introduced called the risk variable. This will reduce the influence of the weights on the overall evaluation of similarity and will also consider the not always positive effect of the solution of the selected case on our problem. This will make the retrieval stage better and more realistic. Keywords: CBR, attribute risk, similarity measure, fuzzy inference system 1 INTRODUCTION Case-based reasoning (CBR) provides a methodology for reasoning and learning and there has been much published research on the subject (Jacek, 2001; Kolodmer, 1993; Roth-Berghofer, 2004). This methodology consists in using previous experiences (called cases) and in adapting previous solutions to solve new problems (i.e. it recalls old problems in order to obtain information about current problems). In a CBR system, previous cases are stored in a case base and are characterized by a set of predefined attributes. When a new problem is encountered, the CBR system works through the following steps: Retrieve similar cases to the new problem, reuse a solution suggested by a similar case, Revise the solution to fit the new problem, Retain the new solution once it has been confirmed. Of these steps, case retrieval is extremely important for the success of a CBR system. Although case-based reasoning is useful for various types of problems and domains (G. O. Yeh, 2001; Schmidt, 2001; Sadek, 2001; Long, 1995), it is not always the most suitable methodology to use. Below, we shall summarise some of the advantages and disadvantages of CBR. Advantages: 1 Since CBR reduces the knowledge acquisition task, the human expert is then released from the task of giving training data.