.png){kind=icon}
.png){kind=icon}
Return
A MULTICATEGORY CLASSIFICATION MODEL WITH REGULARIZED PAIRWISE COMPARISON
.png){kind=icon}
PRE
AI
DOI：10.23055/ijietap.2026.33.2.10891.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multicategory classification poses a significant challenge in machine learning because real-world problems often involve multiple classes, while many algorithms are tailored for binary classification. Among the methods for extending binary classifiers to multicategory classification problems, two representative approaches are the One-versus-The-Rest (OVT) and One-versus-One (OVO) methods. In this paper, we demonstrate that the OVT method can encounter masking issues in specific scenarios. To mitigate this, we propose a new algorithm based on the OVO framework. Specifically, our method integrates the OVO approach with the idea of the Mallows-Bradley-Terry model to estimate class probabilities by optimizing a single penalized pseudo-likelihood objective. This approach allows for the simultaneous estimation and regularization of all pairwise parameters, significantly reducing computational overhead. Experimental results and real data analysis indicate that the proposed method surpasses the OVT in prediction performance and is substantially more robust against the masking problem.
Keywords:
Multicategory Classification
Regularization
One-Versus-One
One-Versus-The Rest
Masking Problem
Journal
I
IF:
1
Papers:
40
Citations:
561
Organization
No organization information available