WEIGHT ANALYSIS AND OPTIMIZATION IN FUZZY MODELING M.G. TSIPOURAS, T.P. EXARCHOS, D.I. FOTIADIS Unit of Medical Technology and Intelligent Information Systems, Department of Computer Science, University of Ioannina, GR 45110 Ioannina, GREECE In this paper we propose the use of a set of weights in fuzzy modelling, the class weights, which are assigned to each class of a classification problem. We automatically generate a fuzzy model, using a three-stage methodology: (i) generation of a crisp model from a decision tree, induced from the data, (ii) transformation of the crisp model into a fuzzy one, and (iii) optimization of the fuzzy model’s parameters. Based on this methodology, the generated fuzzy model includes the Θ f parameters, which are all the parameters included in the sigmoid functions. In addition, local, global and class weights are included, thus the fuzzy model is optimized with respect to all these parameters ( Θ f , local, global and class weights). The class weight introduction, which is a novel approach, grants to the fuzzy model the ability to identify the individual importance of each class and thus more accurately reflect the underlying properties of the classes under examination, in the domain of application. The above described methodology is applied to five known medical classification problems, obtained from the UCI machine learning repository, and the obtained classification accuracy is high. 1. Introduction Fuzzy logic is the extension of the classical crisp (binary) logic into a multivariate form. Fuzzy logic is closer to the human logic, thus being able to deal with real world noisy and imprecise data [1]. Fuzzy models experience several advantages, compared to crisp ones, mainly being more flexible on the decision boundaries, and thus characterized by their higher ability to adjust to a specific domain of application and more accurately reflect its particularities. A fuzzy model can be created by defining an initial crisp model (set of rules) and then fuzzyfing it. This approach is a complex task since several issues must be defined for the fuzzy model to be generated. First, the origin of the rules must be addressed, which determines the philosophy of the method; if expert’s knowledge is used then the generated fuzzy model will be knowledge-based while, if data mining techniques are employed then a data-driven fuzzy model will be generated. In the fuzzyfication step, there are several fundamental features related to the definition of the fuzzy model, such as the fuzzy membership function, the fuzzy operators, the defuzzyfication approach and the use of weights. Following this approach, the generated fuzzy model resembles