Learning Bayesian Networks with Algebraic Differential Evolution

In this paper we introduce DEBN, a novel evolutionary algorithm for learning the structure of a Bayesian Network. DEBN is an instantiation of the Algebraic Differential Evolution which is designed and applied to a particular (product) group whose elements encode all the Bayesian Networks of a given set of random variables. DEBN has been experimentally investigated on a set of standard benchmarks and its effectiveness is compared with BFO-B, a recent and effective bacterial foraging algorithm for Bayesian Network learning. The experimental results show that DEBN largely outperforms BFO-B, thus validating our algebraic approach as a viable solution for learning Bayesian Networks.