Fuzzy Broad Neuroevolution Networks via Multiobjective Evolutionary Algorithms: Balancing Structural Simplification and Performance

Dynamic fuzzy broad learning system (DFBLS) is a fuzzy neural network based on the TSK fuzzy system and broad learning (BL). DFBLS possesses excellent model interpretability and efficient predictive performance. However, due to the phenomenon of rule explosion and the redundancy of network nodes, ba...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on instrumentation and measurement Vol. 74; pp. 1 - 10
Main Authors: Zhao, Huimin, Wu, Yandong, Deng, Wu
Format: Journal Article
Language:English
Published: New York IEEE 2025
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects:
Accuracy
Balancing
Complexity-accuracy trade-off
Computational modeling
Datasets
Evolutionary algorithms
Evolutionary computation
Feature extraction
fuzzy broad networks
Fuzzy logic
fuzzy rule (FR)-based classification systems
Fuzzy sets
Genetic algorithms
Learning systems
Linear programming
Machine learning
multiobjective evolutionary network framework
Multiple objective analysis
Neural networks
Optimization
Predictions
Predictive models
Redundancy
Training
ISSN:0018-9456, 1557-9662
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Dynamic fuzzy broad learning system (DFBLS) is a fuzzy neural network based on the TSK fuzzy system and broad learning (BL). DFBLS possesses excellent model interpretability and efficient predictive performance. However, due to the phenomenon of rule explosion and the redundancy of network nodes, balancing network structure and performance has become a challenge in the construction of the DFBLS. Therefore, for the best balance between model predictive performance and network simplicity, a fuzzy broad neuroevolutionary network via multiobjective evolutionary algorithms (EAs) was developed in this article. First, an effective genetic encoding strategy was designed to represent the feature node building blocks and network connectivity relationships. The incremental mechanism and the randomly connected network relationships in the ILFR structure are replaced by the evolutionary framework. Second, a multiobjective optimization problem model is constructed with the optimization objectives of prediction performance and minimal network structure, and the corresponding objective functions are proposed. Finally, a self-adaptive mutation strategy with scaling genes is proposed for NSGA-II to optimize the accuracy and structure of the neural networks. The experiments demonstrate that SGNSGA-DFBLS achieves the best hypervolume (HV) values in 7 out of 9 public datasets. Its performance on the Air and pm2.5 datasets is either superior to or on par with other recently proposed models. SGNSGA-DFBLS can achieve high test accuracy with fewer fuzzy rules (FRs) and a compact network structure when constructing fuzzy broad network models.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0018-9456
1557-9662
DOI:10.1109/TIM.2024.3485438