Reinforced fuzzy neural networks based on maximum entropy clustering and conjugate gradient method

This paper presents a novel fuzzy clustering-based Takagi–Sugeno–Kang (TSK) neural network for classification tasks, aiming to enhance the accuracy of existing models through a more effective design approach. Unlike previous fuzzy neural networks (FNNs) models, our classifier leverages the maximum e...

Full description

Saved in:
Bibliographic Details
Published in:Engineering applications of artificial intelligence Vol. 142; p. 109909
Main Authors: Dong, Qingmei, Fan, Qinwei, Xing, Zhiwei
Format: Journal Article
Language:English
Published: Elsevier Ltd 15.02.2025
Subjects:
Conjugate gradient method
Fuzzy clustering
Intelligent systems
L2 regularization
Takagi–Sugeno–Kang neural network
Conjugate gradient method
Fuzzy clustering
Takagi–Sugeno–Kang neural network
L2 regularization
Intelligent systems
ISSN:0952-1976
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This paper presents a novel fuzzy clustering-based Takagi–Sugeno–Kang (TSK) neural network for classification tasks, aiming to enhance the accuracy of existing models through a more effective design approach. Unlike previous fuzzy neural networks (FNNs) models, our classifier leverages the maximum entropy clustering algorithm to establish the first layer structure, while updating the hidden layer weights to address limitations of traditional clustering methods. Additionally, to alleviate computational burden, we employ the conjugate gradient method to adaptively adjust connection parameters of the output layer, employing a softmax function to represent probability distribution. Furthermore, we introduce the cross-entropy error function as the objective function and include an L2 regularization term to mitigate multicollinearity, thereby preventing overfitting and enhancing generalization performance. Experimental evaluations on two-dimensional and other machine learning datasets demonstrate the superiority of our model through comprehensive comparative analyses, highlighting the impact of various parameters on classifier performance. By implementing this new model, we explore the potential of combining fuzzy clustering with TSK neural networks for applications in the fields of pattern recognition and optimization of intelligent systems. [Display omitted]
ISSN:0952-1976
DOI:10.1016/j.engappai.2024.109909