Stochastic configuration networks with robust supervised least squares regression

Stochastic Configuration Networks (SCNs) are widely used in regression modeling to fit data distributions due to their fast convergence and powerful learning capabilities. However, in practical industrial applications, labeled outliers in the dataset inevitably appear, which can destroy the internal...

Full description

Saved in:
Bibliographic Details
Published in:Neurocomputing (Amsterdam) Vol. 647; p. 130548
Main Authors: Chu, Fei, Sun, Zihang, Wang, Yu, Zhang, Yong
Format: Journal Article
Language:English
Published: Elsevier B.V 28.09.2025
Subjects:
Least squares regression
Randomized algorithms
Robust regression
Stochastic configuration network
Randomized algorithms
Robust regression
Stochastic configuration network
Least squares regression
ISSN:0925-2312
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Stochastic Configuration Networks (SCNs) are widely used in regression modeling to fit data distributions due to their fast convergence and powerful learning capabilities. However, in practical industrial applications, labeled outliers in the dataset inevitably appear, which can destroy the internal structure and connection of the data, and not only seriously reduce the prediction accuracy of SCNs, but may also increase the risk of overfitting. To cope with this problem, this paper proposes a new algorithm - Robust Supervised Stochastic Configuration Networks (RSSCNs). In this paper, firstly, the loss function is constructed based on the concave function of the error, which extends the applicability of the robust supervised least squares regression. Second, robust supervised least squares regression is introduced into the SCN algorithm for the first time, which enhances the robustness of the model and reduces the risk of model overfitting. Finally, the inequality constraint of the SCN algorithm is reconstructed according to the loss function, which optimizes the parameter selection and accelerates the convergence of the model. In the experimental section, RSSCNs are applied to a function fitting experiment, several benchmark datasets, and an industrial dataset from a dense medium coal preparation process, and the performance is compared with other algorithms. The experimental results show that for these datasets, RSSCNs perform favorably in terms of accuracy and robustness, validating the effectiveness of the algorithm.
ISSN:0925-2312
DOI:10.1016/j.neucom.2025.130548