An Improved Genetic Algorithm for Optimal Tree Decomposition of Bayesian Networks

In this paper, we present a novel triangulation heuristic and a new genetic algorithm to solve the problem of optimal tree decomposition of Bayesian networks. The heuristic, named MinFillWeight, aims to select variables minimizing the multiplication of the weights on nodes of fill-in edges. The genetic algorithm, named IDHGA, employs a new order-reserving crossover operator and a mutation operator based on triangulation heuristics. Moreover, IDHGA utilizes population diversity to identify stagnation and convergence as well as to guide the search procedure. Experiments on representative benchmarks show that IDHGA utilizing MinFillWeight heuristic posses better performance and robustness than other swarm intelligence methods.