A hybrid method for learning Bayesian networks based on ant colony optimization

As a powerful formalism, Bayesian networks play an increasingly important role in the Uncertainty Field. This paper proposes a hybrid method to discover the knowledge represented in Bayesian networks. The hybrid method combines dependency analysis, ant colony optimization (ACO), and the simulated annealing strategy. Firstly, the new method uses order-0 independence tests with a self-adjusting threshold value to reduce the size of the search space, so that the search process takes less time to find the near-optimal solution. Secondly, better Bayesian network models are generated by using an improved ACO algorithm, where a new heuristic function is introduced to further enhance the search effectiveness and efficiency. Finally, an optimization scheme based on simulated annealing is employed to improve the optimization efficiency in the stochastic search process of ants. In a number of experiments and comparisons, the hybrid method outperforms the original ACO-B which uses ACO and some other network learning algorithms.