A Development of Fuzzy Rule-based Regression Models through Using Decision Trees
Xiubin Zhu,Xingchen Hu,Lan Yang,Witold Pedrycz,Zhiwu Li
DOI: https://doi.org/10.1109/tfuzz.2024.3365572
IF: 12.253
2024-01-01
IEEE Transactions on Fuzzy Systems
Abstract:This paper presents a design and realization of fuzzy rule-based regression models based on standard decision trees. A two-phase design of rule-based model is offered in this study to provide a good alternative to cope with high dimensional data. We first build a standard decision tree on the basis of variables in order to discover homogeneous subsets of the data. Subsequently, a collection of fuzzy rules is induced by the decision tree with the aim of reflecting the underlying phenomenon. The calculation of membership degrees and the refinement of fuzzy rules on the basis of data located in each partition exhibit a substantial level of originality and innovation. The introduction of fuzziness into decision rules helps to characterize and quantify the continuous change of output values near the boundary areas. The constructed fuzzy rules could efficiently handle the ambiguity and vagueness in the experimental evidence and offer accurate characterization of the nonlinearities of the input-output relationships. The developed fuzzy models could achieve much higher prediction accuracy in comparison with traditional decision trees of the same size and fuzzy rule-based models with the same number of rules. Another advantage of the proposed methodology comes with the evident readability of the formed fuzzy rules. A series of experiments is reported to demonstrate the superiority of the proposed architecture of fuzzy rule-based models over traditional fuzzy rule-based models and decision trees.
computer science, artificial intelligence,engineering, electrical & electronic