Weighting fuzzy classification rules using receiver operating characteristics (ROC) analysis

In fuzzy rule-based classification systems, rule weight has often been used to improve the classification accuracy. In past research, a number of heuristic methods for rule weight specification have been proposed. In this paper, a method of fuzzy rule weight specification using Receiver Operating Characteristic (ROC) analysis is proposed. In order to specify the weight of a fuzzy rule, using 2-class ROC analysis, the threshold that the rule achieves its maximum accuracy is found. This threshold is used as the weight of the rule. The proposed method is compared with existing ones through computer simulations on some well-known classification problems with continuous attributes. Simulation results show that the proposed method performs better than existing methods of fuzzy rule weight specification.