Designing an optimal evolutionary fuzzy decision tree for data mining
Hung-Ming Chen, Shinn‐Ying Ho · 2001
The amount of digital data processed by computers grows extremely fast. Therefore, how to discover useful knowledge from large digital data has become a very important issue. Decision trees like ID3 and C4.5 are efficient classifiers in generating classification rules from large training patterns. Besides, decision tree classifiers can directly generate linguistic if-then rules which are human understandable knowledge. The design of an optimal evoltionary fuzzy decision tree is proposed in this paper. Fuzzy decision tree integrates the flexibility of fuzzy sets and comprehensibility of decision trees. In order to generate a more compact fuzzy decision tree, we use flexible trapezoid fuzzy sets to define membership functions. The parameters of membership functions are optimized by a powerful intelligent genetic algorithm. Furthermore, additional control genes in a chromosome are used to perform feature selection and redundant fuzzy set deletion simultaneously. The performance of the proposed fuzzy decision tree is superior to conventional decision trees and the existing fuzzy rule-based approaches in terms of both classification rate and number of fuzzy rules.