Fuzzy rule mining by fuzzy decision tree induction based on fuzzy feature subset

Fuzzy decision tree (FDT) induction can be used to generate and mine weighted fuzzy production rules (WFPRs) and the representation power of WFPRs can be enhanced by including several knowledge parameters such as weight and certainty factor. So far, the heuristic used in FDT generation is the entropy-based heuristic which can learn a set of fuzzy rules from examples with the same type of attributes, but it cannot handle data with mixed attributes and cannot capture knowledge parameters of WFPRs. This paper proposes a new heuristic which can not only generate and mine a set of WFPRs from data with mixed attributes with better performance but also capture the knowledge parameters of fuzzy rules. First, a uniform fuzzy representation of training examples with mixed attributes is proposed. Then a new heuristic for generating a FDT to which FPRs are extracted is given by using the fuzzy feature subset. After that, a matching algorithm for classifying an unknown object is presented. Our matching algorithm is shown to have less ambiguity when comparing with another method. Finally, the advantages of our proposed methodology are verified and compared with the Fuzzy ID3 on real-world data.