Fuzzy rule extraction from ID3-type decision trees for real data

This paper proposes a method to construct a fuzzy rule-based classifier system from an ID3-type decision tree (DT) for real data. The three major steps are rule extraction, gradient descent tuning of the rule-base, and performance-based pruning of the rule-base. Pruning removes all rules which cannot meet a certain level of performance. To test our scheme, we have used the DT generated by RIB3, an ID3-type classifier for real data. In this process, we made some improvements of RID3 to get a tree with less redundancy and hence a smaller rule-base. The rule-base is tested on several data sets and is found to demonstrate an excellent performance. Results obtained by the proposed scheme are consistently better than C4.5 across several data sets.