A proposed method for learning rule weights in fuzzy rule-based classification systems

In fuzzy rule-based classification systems (FRBCSs), rule weighting has often been used as a simple mechanism to tune the classifier. In past research, a number of heuristic rule weight specification methods have been proposed for this purpose. A learning algorithm based on reward and punishment has also been proposed to adjust the weights of each fuzzy rule in the rule-base. In this paper, a new method of learning rule weight in FRBCSs is proposed. The method can be used when single winner or weighted vote methods of reasoning is used. Compared with reward and punishment scheme, the proposed method is much faster and more effective. Another advantage of the proposed method is that, during the learning process, redundant rules are removed (i.e., by setting their weights to zero). The final rule-base usually contains much fewer rules than the initial one. This feature is very useful since a compact rule-base is usually desired from the point of efficiency and interpretability. A number of UCI data sets are used to assess the performance of the proposed method in comparison with reward and punishment scheme.