A staged approach for generation and compression of fuzzy classification rules

A staged approach to identify a compact fuzzy classification rule base from numerical data is presented. First, the fuzzy rules are generated by adaptively clustering the input data and defining a relationship between cluster membership values and class labels. Then, the classification accuracy of the resulting fuzzy rules is enhanced by training a neuro-fuzzy network used to model the fuzzy classifier. Finally, the interpretability of the resulting fuzzy classifier is improved via a compression of the fuzzy rule base. Two well known data classification problems are considered to asses the validity of the approach.