Copula Estimation of Distribution Algorithm Sampling from Clayton Copula

Estimation of Distribution Algorithms (EDAs) are implemented mainly by the three steps: selecting the promising subset from the current population, modeling the distribution of the selected population and sampling from the estimated model. Modeling and sampling are key steps of EDAs. They are also research topic of copula theory to represent the multivariate joint distribution by a copula and the one-dimensional marginal distributions and to sample from copula. A kind of EDA modeling and sampling from Clayton copula is proposed in this paper. The joint distribution of variables is not estimated directly. But it is described with Clayton copula and the one-dimensional margins. Only the margins are estimated in the proposed EDAs. And a sampling method from Clayton copula is used in the algorithm. The next generation is produced according to the inverses of the margins and the generated samples from Clayton copula. The experiments on the test functions show that the algorithm is valid and efficient.