Hybridisation of evolutionary programming and machine learning with k-nearest neighbor estimation

Evolutionary programming (EP) focus on the search step size which decides the ability of escaping local minima, however does not touch the issue of search in promising region. Estimation of distribution algorithms (EDAs) focus on where the promising region is, however have less consideration about behavior of each individual in solution search algorithms. Since the basic ideas of EP and EDAs are quite different, it is possible to make them reinforce each other. In this paper, we present a hybrid evolutionary framework to make use of both the ideas of EP and EDAs through introducing a mini estimation operator into EP´s search cycle. Unlike previous EDAs that use probability density function (PDF), the estimation mechanism used in the proposed framework is the k-nearest neighbor estimation which can perform better with relative small amount of training samples. Our experimental results have shown that the incorporation of machine learning techniques, such as k-nearest neighbor estimation, can improve the performance of evolutionary optimisation algorithms for a large number of benchmark functions.