B1432
Title: Kriging with continuous and categorical inputs in R
Authors: Dominik Kirchhoff - Dortmund University of Applied Sciences and Arts (Germany) [presenting]
Abstract: The implementation of extended Kriging models for mixed continuous and categorical input variables in R is considered. Kriging models (also known as Gaussian Stochastic Process models) are an important tool in metamodel-based optimization as they provide not only fast predictions for expensive processes (e.g., simulations), but also an uncertainty estimator of these predictions. The original Kriging model, however, can only cope with purely continuous input variables. First extensions exist to incorporate also categorical variables. We consider three approaches called Exchangeable Correlation, Multiplicative Correlation, and Unrestrictive Hypersphere-based Correlation, which are different in terms of their flexibility and computational effort. We also implement two new distance-based methods and discuss the advantages and disadvantages of each of the methods.
