Author(s)

E R Hruschka & N F F Ebecken

Abstract

A clustering genetic algorithm for extracting rules from multilayer perceptions trained in classification problems E. R. Hruschka1 & N.F.F. Ebecken2 1 Universidade Tuiuti do Paranci, Brazil. 2 COPPE /Universidade Federal do Rio de Janeiro, Brazil. Abstract This paper deals with the task of using supervised neural networks in data mining applications. The proposed methodology makes use of a clustering genetic algorithm, which is applied in the hidden units activation space in order to extract rules from multilayer perceptions trained in classification problems. We illustrate the proposed method by means of two examples: Iris Plants Database and Meteorological dataset. 1 Introduction Multilayer perceptions (MP) adjust their internal parameters performing vector mappings from the input to the output space. Although they may achieve high accuracy of classification, the knowledge acquired by such neural networks is usually incomprehensible for humans [1]. This fact can be a major obstacle in domains such as data mining where it is important to have symbolic rules or other forms of knowledge structure [2]. Therefore, many methods have been developed to get explicit knowledge from these models. One observes that, fundamentally, the knowledge acquired by a neural network is codified on the connection weights that, in turn, determine the activation function values. Thus, the knowledge acquisition process from supervised neural networks implies the use of algorithms based either on the connection weight values or on the hidden unit activation values. The algorithms designed to perform this task are usually called algorithms for rule extraction from neural networks This paper describes the application of a Clustering Genetic Algorithm (CGA) for rule extraction horn MP. The rule extraction algorithm basically

Keywords