Response surface analysis as a technique for the visualization of linear models and data

Abstract

Response surface analysis aims of exploring the behavior of a response function in some region. Some authors (see e.g. [BRU] and [DEV]), have used this approach to get "better" interpretations of the behavior of linear models over the convex hull of a set of data points. This has been done by solving linear programs with feasible regions confined to the convex hull of the set of data points. This gives rise to graphical representations that can be used as and aid to the modeling process and the interpretation of a specific linear model. One of the results of this methodology is that the influence of specific variables of interest can easily be deduced from these graphical representations. In this paper we look at the feasibility of creating decision support systems based on this technology in an interactive high performance computer environment.