Bayesian Optimization Algorithm for Learning Structure of Dynamic Bayesian Networks from Incomplete Data

Abstract

An algorithm based on Bayesian optimization algorithm (BOA), BOA-DBN, is proposed to learn the structure of DBN from incomplete databases. The algorithm takes fitness function based on expectation, which can convert incomplete data into complete data utilizing current best learned dynamic Bayesian network in evolutionary process. BOA generates a population of strings for the next generation, which tends to develop according to the optimization direction under the fitness function. Thus DBNs can be learned by using two Bayesian networks, prior network and transition network, to reduce the computational complexity. Encoding is presented, and genetic operators which provides guarantee of convergence are designed. Experimental results show that, given a missing data set, this algorithm can learn a DBN very close to the generative model and at the same time, enjoy the tend to converge at global optima due to BOA.