Approximation Models in Robust Optimization Over Time - An Experimental Study

Robust optimization over time (ROOT) is a novel research topic in the field of evolutionary dynamic optimization, in which only just there has been few advances. Specifically, one of the less developed topic is how to define an approximation model (AM) that suitably characterizes the past environments. For a ROOT algorithm this is a key issue that affects its performance, because AMs are used for estimating robustness (e.g by predicting fitness in future environments). In this work we studied the role of several approximation models (AM) for characterizing past environments in the context of robust optimization over time (ROOT). We have conducted several computational experiments in order to analyze the effects of different AMs in the algorithm performance over continuous ROOT problems. We observed that the Support Vector Regression model with Laplacian Kernel allows the algorithm to achieve a significantly, high performance.