Preselection via classification: a case study on global optimisation

Abstract

In evolutionary optimisation, the preselection aims to choose promising solutions from a set of candidates for the fitness evaluation. It is usually based on the approximated fitness values, which are not necessary in many cases because we are usually interested in whether a candidate is promising or not instead of how promising it is. Actually, the preselection can be regarded as a classification process, i.e., to assign each candidate solution a label (+1 if promising or –1 otherwise). To this end, this paper proposes a classification based preselection (CPS) strategy and applies it to evolutionary optimisation. Systematic experiments are conducted to study the performance of CPS and the experimental results suggest that it can significantly improve the performance of some state-of-the-art evolutionary algorithms on most of the given test instances.