Machine Learning Alternatives to Response Surface Models

In the Design of Experiments, we seek to relate response variables explanatory factors. Response Surface methodology (RSM) approximates relation between output and a polynomial transform using linear model. Some researchers have tried adjust other types models, mainly nonlinear nonparametric. We present large panel Machine Learning approaches that may be good alternatives classical RSM approximation. The state art such is given, including classification regression trees, ensemble methods, support vector machines, neural networks also direct multi-output approaches. survey subject illustrate use ten simulations real case. our simulations, underlying model in factors for one others. focus on advantages disadvantages different show how their hyperparameters tuned. Our even when linear, approach outperformed by network multivariate model, any sample size (<50) much more very small samples (15 or 20). When nonlinear, most machine learning (n ≤ 30).