MOKBL plus MOMs: An interpretable multi-objective evolutionary fuzzy system for learning high-dimensional regression data

摘要

This work presents a multi-objective evolutionary linguistic fuzzy system that addresses regression problems, especially those that are dimensional and scalable. A multi-objective knowledge base learning (MOKBL) is developed in the first stage of this model. MOKBL learns the most relevant and least redundant features by considering the desirability of the components of the fuzzy system. At the same time as feature selection, MOKBL slightly tunes the membership functions to provide greater initial adaptation of the fuzzy rule-based system components. In the second stage, multi-objective modifications (MOMs) are organized to modify the generated fuzzy system and to perform post-processing tasks. MOMs more finely tune the membership functions and prune additional rules. The newly proposed rule pruning method can eliminate weak rules from the rule base using the concepts of support and confidence. The membership functions tuning process is accomplished using the tasks of core displacement and width alteration of the symmetric functions. MOKBL+MOMs and its stages were validated using 28 real-world datasets and compared with two state-of-the-art regression solutions through non-parametric statistical tests. The experimental results confirmed the effectiveness of MOKBL+MOMs in terms of interpretability (complexity), accuracy, and time. (C) 2019 Elsevier Inc. All rights reserved.