Generalized eigenvalue proximal support vector regressor

Reshma Khemchandani, Anuj Karpatne, Suresh Chandra

Research output: Contribution to journalArticlepeer-review

11 Scopus citations

Abstract

In this paper, we propose a new non-parallel plane based regressor termed as Generalized Eigenvalue Proximal Support Vector Regressor (GEPSVR). The GEPSVR formulation is in the spirit of non-parallel plane proximal SVMs via generalized eigenvalues and is obtained by solving two generalized eigenvalue problems. Further, an improvement over GEPSVR is proposed that employs a regularization technique, similar to the one proposed in Guarracino, Cifarelli, Seref, and Pardalos (2007), which requires the solution of a single regularized eigenvalue problem only. This regressor has been termed as Regularized GEPSVR (ReGEPSVR). On several benchmark datasets and artificially generated datasets, ReGEPSVR is not only fast, but also shows good generalization when compared with other regression algorithms. It also finds its application in financial time-series forecasting, as shown over financial datasets.

Original languageEnglish (US)
Pages (from-to)13136-13142
Number of pages7
JournalExpert Systems with Applications
Volume38
Issue number10
DOIs
StatePublished - Sep 15 2011

Keywords

  • -insensitive bound
  • Generalized eigenvalues
  • Regression
  • Regularization
  • Support vector machines

Fingerprint Dive into the research topics of 'Generalized eigenvalue proximal support vector regressor'. Together they form a unique fingerprint.

Cite this