A Method for Dimension Reduction in Quadratic Classification Problems Articles uri icon

publication date

  • September 2008

start page

  • 572

end page

  • 589

issue

  • 3

volume

  • 17

international standard serial number (ISSN)

  • 1061-8600

electronic international standard serial number (EISSN)

  • 1537-2715

abstract

  • This article presents a dimension-reduction method in quadratic discriminant analysis (QDA). The procedure is inspired by the geometric relation that exists between the subspaces used in sliced inverse regression (SIR) and sliced average variance estimation (SAVE). A new set of directions is constructed to improve the properties of the directions associated with the eigenvectors of the matrices usually considered for dimension reduction in QDA. Illustrative examples of application with real and simulated data are discussed.