MARS approach to response surface models

In this study, the approximation of the multivariate adaptive regression splines (MARS) is proposed at the stage of modeling in second order design in response surface methodology. It is mentioned that what kind of factorial designs and datasets can be used for this approximation and also how the modeling stage can be made is explained. In the application part of the study, the pollution of heavy metals coming from $3^2$ design for both soil and road dust datasets which are collected from a specific area in Eskisehir are classified by the means of regression trees and is modeled by the use of response surface methodology and MARS. The results are evaluated by statistical tests and by some criteria. Two programs, based on regression trees for a univarite case and a bivariate case, are generated using R Software. Two different programs for a univarite case and a bivariate case to model the pollution of heavy metal data by MARS are generated in R Software.