Bayesian network learning on algorithm based on ant colony optimization

Accordering to the hybrid Bayesian networks learning algorithms which are easy to narrow the search space and fall into local optimum, a Bayesian network learning algorithm based on ant colony optimization is proposed. Firstly, this paper applies max-min parents and children (MMPC) to construct the framework of the undirected network, and then uses ant colony optimization to score-search, by balancing the "exploitation" and "exploration" to repair the search space and determine the direction of edges in the network. Finally applying the algorithm to learn a logical alarm reduction mechanism (ALARM) network shows that it reduces the number of missing edges, and gets closer to the real structure of Bayesian network.