Using Boosting Techniques to Improve the Performance of Fuzzy Classification Systems

Abstract

This paper proposes a boosting algorithm of fuzzy rule-based systems for pattern classification. In the proposed algorithm, several fuzzy rule-based classifi- cation systems are simultaneously used for classifying an input pattern. Those fuzzy rule-based classification systems are generated from different subsets of training patterns. A subset of training patterns for generating each fuzzy rule-based classi- fication system is generated depending on the performance of previously generated fuzzy rule-based classification systems. It is expected that the proposed algorithm performs well on both training data and test data. The performance of the proposed algorithm is shown in computer simulations on two real-world pattern classification problems such as the iris data set and the appendicitis data set.