Designing Cost-sensitive Fuzzy Classification Systems Using Rule-weight
Mansoor Zolghadri Jahromi, Mohammad Reza Moosavi · 2011
In the field of pattern classification, we often encounter problems that class-to-class misclassification costs are not the same. For example, in the medical domain, misclassifying a patient as normal is often much more costly than misclassifying a normal as patient. Our aim in this paper is to propose a method of designing fuzzy rule-based classification systems to tackle this problem. We use rule- weight as a simple mechanism to tune the rule-base. Assuming that class-to-class misclassification costs are known, we propose a learning algorithm that attempts to minimize the total cost of the classifier on train data (i.e., instead of minimizing the error-rate). Using a number of UCI datasets we show that the method is quite effective in reducing the average cost of the classifier on test data.