.png){kind=icon}
Return
An efficient Universum-based twin parametric-margin random vector functional link classifier
.png){kind=icon}
PRE
AI
DOI：10.1016/j.compeleceng.2026.111285.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The Twin Random vector functional link (TRVFL) is a popular and effective model that can be used to solve real-world regression and classification issues. However, TRVFL does not consider heteroscedasticity in real-world applications. Therefore, its performance suffers severely when there are outliers in the dataset. TRVFL generally works with the assumption of a consistent noise level across the complete domain of the training data. Relaxation of this assumption is possible if the functional dependencies of the noise are known a priori. The hypothesis of a uniform noise model is frequently not satisfied in real life scenario. Consider, for example, the case of heteroscedastic noise, where the noise level is a function of location. Also, TRVFL losses its sparsity and it is sensitive to class imbalance. To address these limitations of TRVFL, the twin-parametric margin concept is induced in TRVFL. To enhance the overall classification performance of the TRVFL network and boost its classification efficiency on data sets that contain outliers, this paper introduces a novel Universum twin parametric-margin RVFL (UTPMRVFL) for binary classification. Firstly, to solve two small RVFL-type issues, the proposed UTPMRVFL builds two non-parallel parametric hyperplanes. The model's performance is improved by incorporating prior information i.e., Universum data. The incorporation of prior knowledge about the data leads to a better optimized classifier. To evaluate the effectiveness of the suggested model, its classification performance has been compared with Random vector functional link (RVFL), Twin RVFL(TRVFL), Angle based TRVFL (ATRVFL), Universum Twin SVM (UTSVM), Twin parametric-margin SVM (TPMSVM), ensemble deep learning-based RVFL network (edRVFL) and 1-Norm Universum Twin RVFL (UTRVFL1-Norm). Empirical evidence shows that the proposed technique is both efficient and sustainable. The proposed UTPMRVFL attained a classification accuracy of 94.65% and a maximum F1-score of 0.810 and Area Under Curve (AUC) of 0.905 when evaluated on real-world datasets and leaf datasets. Experimental investigation reveals that the proposed UTPMRVFL performs better than, RVFL, TRVFL, TPMSVM, UTSVM, edRVFL, UTRVFL1-Norm and ATRVFL.
Journal
C
IF:
4.9
Papers:
6.7K
Citations:
1.3W
Organization
View more organizations
Cited Papers
No cited papers available