Fuzzy classification with reject options by fuzzy if-then rules

Abstract

We discuss fuzzy rule-based classification systems with reject options. In such systems, classification of new patterns close to class boundaries is usually rejected. The rejection of such doubtful patterns can reduce misclassification rates (i.e. improve the reliability of fuzzy rule-based classification systems). An exception handling is applied to each of the rejected patterns. We first describe three fuzzy reasoning methods for pattern classification problems. Two methods are based on fuzzy if-then rules with single consequent class, and the other is based on those with multiple consequent classes. Next, reject options are introduced to each fuzzy reasoning method. Then, the performance of the fuzzy rule-based classification systems with a reject option is examined by computer simulations.