Some Problems in Orthogonal Distance and Non-Orthogonal Distance Regression
Abstract
Of interest here is the problem of fitting a curve or surface to given data by minimizing some norm of the distances from the points to the surface. These distances may be measured orthogonally to the surface, giving orthogonal distance regression, and for this problem, the least squares norm has attracted most attention. Here we will look at two other important criteria, the iota(sub 1)i norm and the Chebyshev norm. The former is of value when the data contain wild points, the latter in the context of accept/reject criteria. There are however circumstances when it is not appropriate to force the distances to be orthogonal, and two possibilities of this are also considered. The first arises when the distances are aligned with certain fixed directions, and the second when angular information is available about the measured data points. For the least squares norm, we will consider some algorithmic developments for these problems.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2001
- Accession Number
- ADP013740
Entities
People
- G. A. Watson
Organizations
- University of Dundee