Input selection in fuzzy rule-based classification systems

Real-world pattern classification problems usually involve many attributes. In such a pattern classification problem, all attributes are not always necessary for the classification task. That is, when we design a pattern classification system, some attributes may be removable with no deterioration of its performance. The main aim of this paper is to describe how the number of attributes can be reduced when we design a fuzzy rule-based classification system. We use a simple stepwise input selection mechanism: first the most important two attributes are selected by the examination of all combinations, then the next best attribute is added sequentially. By computer simulations on well-known real-world test problems with many continuous attributes, the performance of the fuzzy rule-based classification system designed by the input selection mechanism is examined. Simulation results clearly show that a small number of selected attributes have high classification ability for many real-world test problems