Incremental Learning of Bayesian Networks with Hidden Variables

摘要

In recent years, there has been a growing interest in learning Bayesian network structure from data. However, most of the research work has concentrated on batch learning methods. While in many applications, such as monitoring systems, embedded systems and data mining, it is very important and meaningful to learn Bayesian network structure incrementally. But incremental learning of Bayesian network structure is still an open problem, especially in the presence of incomplete data and hidden variables. In this paper, an incremental learning method based on evolutionary computing, IEMA, is put forward. IEMA combines the EM algorithm with evolutionary algorithm organically, and transforms the incomplete data to complete data by EM algorithm and then evolves network structures by the evolutionary algorithm. Through introducing a new mutation operator and expanding the crossover operator, IEMA could learn and evolve Bayesian networks with hidden variables. Thus, IEMA can learn Bayesian network structures incrementally not only from complete data, but also in the presence of missing data and hidden variables. The results of the experiments verified the validity of IEMA. In terms of storage cost, IEMA is comparable with the incremental learning method of Friedman et al, while it is more accurate.