V. Blanco, J. Puerto Albandoz, R. Salmerón
This paper presents a family of new methods for estimating the coefficients in multiple linear regression models. The novelty consists in considering distance-based residuals instead of the usual vertical distance and on the use of different forms of aggregation criteria for those residuals. The most popular methods found in the specialized literature can be cast within this family as particular choices of the residuals and the aggregation criteria. Mathematical programming formulations for these methods are stated and some interesting cases are analyzed. It is also proposed a generalized coefficient of determination to deal with the goodness of fitting for this family of methods. A series of illustrative examples and extensive computational experiments implemented in \r{} are provided to show the performances of some of the proposed methods.
Keywords: Continuous location, mixed integer second order cone programming
Scheduled
T4 Location 3
October 1, 2015 5:00 PM
Salón de actos
