An optimization method of selecting parameters in building fuzzy decision trees and the application in customer forfeit crisis

Compared with conventional decision tree (CDT), fuzzy decision tree (FDT) inductive learning algorithms are more powerful and practical to handle with ambiguities in classification problems. A parameter, named significant level (SL)alpha, plays an important role in the entire process of building FDTs. however this parameter is usually estimated based on users by domain knowledge, personal experience or requirements, which is hard to build a high performance FDT. This paper aims at developing a method to determine an optimal SL value through analysing the relationship between the fuzzy entropy and alpha. On the basis of pointing out the advantage of FDT in solving fuzziness of customer data compared with CDT, the method is applied to analyse the customer forfeit crisis for rising competition ability of enterprises. Experiment is made to prove that FDT, built by the optimal SL, can lead to better classification performance.