Comparing Random Forest and Gaussian Process Modeling in the Gp-demo Algorithm

Abstract

In surrogate-model-based optimization, the selection of an appropriate surrogate model is very important. If solution approximations returned by a surrogate model are accurate and with narrow confidence intervals, an algorithm using this surrogate model needs less exact solution evaluations to obtain results comparable to an algorithm without surrogate models. In this paper we compare two well known modeling techniques, random forest (RF) and Gaussian process (GP) modeling. The comparison includes the approximation accuracy and confidence in the approximations (expressed as the confidence interval width). The results show that GP outperforms RF and that it is more suitable for use in a surrogate-model-based multiobjective evolutionary algorithm.