Two evolutionary methods for learning Bayesian network structures

This paper describes two approaches based on evolutionary algorithms for determining Bayesian networks structures from a database of cases. One major difficulty when tackling the problem of structure learning with evolutionary strategies is to avoid the premature convergence of the population to a local optimum. In this paper, we propose two methods in order to overcome this obstacle. The first method is a hybridization of a genetic algorithm with a tabu search principle whilst the second method consists in the application of a dynamic mutation rate. For both methods, a repair operator based on the mutual information between the variables was defined to ensure the closeness of the genetic operators. Finally, we evaluate the influence of our methods over the search for known networks.