Kernelizing the Proportional Odds Model through the Empirical Kernel Mapping

Research areas:

• Methodology - Machine learning

Year:

2013

Type of Publication:

In Proceedings

Keywords:

Proportional Odds Model, Ordinal Regression, Kernel Trick

Authors:

• Pérez-Ortiz, María
• Gutiérrez, Pedro Antonio
• Cruz-Ramírez, Manuel
• Sánchez-Monedero, Javier
• Hervás-Martínez, César

Volume:

7902

Book title:

12th International Work Conference on Artificial Neural Networks (IWANN2013)

Series:

Lecture Notes on Computer Science (LNCS)

Pages:

270-280

Organization:

Tenerife, Spain

Month:

12th-14th June

ISBN:

978-3-642-38678-7

BibTex:

@conference{KPOMEKM2013, author = "Mar{\'i}a P{\'e}rez-Ortiz and Pedro Antonio Guti{\'e}rrez and Manuel Cruz-Ram{\'i}rez and Javier S{\'a}nchez-Monedero and C{\'e}sar Herv{\'a}s-Mart{\'i}nez", abstract = "The classification of patterns into naturally ordered labels is referred to as ordinal regression. This paper explores the notion of kernel trick and empirical feature space in order to reformulate the most widely used linear ordinal classification algorithm (the Proportional Odds Model or POM) to perform nonlinear decision regions. The proposed method seems to be competitive with other state-of-the-art algorithms and significantly improves the original POM algorithm when using 8 ordinal datasets. Specifically, the capability of the methodology to handle non-linear decision regions has been proven by the use of a non-linearly separable toy dataset. ", booktitle = "12th International Work Conference on Artificial Neural Networks (IWANN2013)", doi = "10.1007/978-3-642-40846-5_47", isbn = "978-3-642-38678-7", keywords = "Proportional Odds Model, Ordinal Regression, Kernel Trick", month = "12th-14th June", organization = "Tenerife, Spain", pages = "270-280", series = "Lecture Notes on Computer Science (LNCS)", title = "{K}ernelizing the {P}roportional {O}dds {M}odel through the {E}mpirical {K}ernel {M}apping", url = "dx.doi.org/10.1007/978-3-642-40846-5_47", volume = "7902", year = "2013", }

Abstract:

The classification of patterns into naturally ordered labels is referred to as ordinal regression. This paper explores the notion of kernel trick and empirical feature space in order to reformulate the most widely used linear ordinal classification algorithm (the Proportional Odds Model or POM) to perform nonlinear decision regions. The proposed method seems to be competitive with other state-of-the-art algorithms and significantly improves the original POM algorithm when using 8 ordinal datasets. Specifically, the capability of the methodology to handle non-linear decision regions has been proven by the use of a non-linearly separable toy dataset.