Multiobjective Evolution of the Explainable Fuzzy Rough Neural Network With Gene Expression Programming

• The fuzzy logic-based neural network usually forms fuzzy rules via multiplying the input membership degrees, which lacks expressiveness and flexibility. In this article, a novel neural network model is designed by integrating the gene expression programming into the interval type-2 fuzzy rough neural network, aiming to generate fuzzy rules with more expressiveness utilizing various logical operators. The network training is regarded as a multiobjective optimization problem through simultaneously considering network precision, explainability, and generalization. Specifically, the network complexity can be minimized to generate concise and few fuzzy rules for improving the network explainability. Inspired by the extreme learning machine and the broad learning system, an enhanced distributed parallel multiobjective evolutionary algorithm is proposed. This evolutionary algorithm can flexibly explore the forms of fuzzy rules, and the weight refinement of the final layer can significantly improve precision and convergence by solving the pseudoinverse. Experimental results show that the proposed multiobjective evolutionary network framework is superior in both effectiveness and explainability.

Ämnesord

NATURVETENSKAP  -- Data- och informationsvetenskap -- Datavetenskap (hsv//swe)

NATURAL SCIENCES  -- Computer and Information Sciences -- Computer Sciences (hsv//eng)

Nyckelord

Fuzzy neural networks

Optimization

Fuzzy logic

Neural networks

Rough sets

Time series analysis

Artificial neural networks

Distributed parallelism

explainability

gene expression programming (GEP)

interval type-2 fuzzy rough neural network (IT2FRNN)

multiobjective evolution

Publikations- och innehållstyp

ref (ämneskategori)

art (ämneskategori)