Reduction of fuzzy systems through open product analysis of genetic algorithm-generated fuzzy rule sets

We explore the reduction of a fuzzy classifier designed to perform a binary classification of tracked or wheeled vehicles based on acoustic data. A genetic algorithm is used to explore the design space of the classifier, with variations performed on the number of antecedents included in the final fuzzy system. Besides the original individual set generated by the GA, we define a subset of it with a small number of antecedents as a filtered set. A novel method of extracting important system components, known as open product analysis, is applied to these two sets, yielding systems that perform well with a small number of antecedents. The fuzzy classifier we reduced performs well using only 20 to 30% of the antecedents that were originally used for classification.