GA-based optimization of fuzzy rule bases for pattern classification

Many problems can be cast as pattern classification problems. Consequently, developing effective classifiers has become an important research area. Various techniques have been proposed to produce classifiers, however many of these appear to the user as “black boxes” which merely give a decision without any additional insight. In this lecture, the focus will be on fuzzy rule-based classification systems which generate simple if-then rules that can thus also be interpreted by the user. Since rule-based classifiers are prone to rule explosion, It will be presented, in particular, optimization approaches to rule base generation that are based on genetic algorithms and a shown to result in a compact yet effective set of rules. In addition, through a simple modification, the resulting classifier can be made cost-sensitive which is in particular useful for applications in medical diagnosis. Example applications will include the classification of gene expression data and the use of classifiers for breast cancer diagnosis.