Handling the incomplete data problem using Bayesian networks

Much of the current research in learning Bayesian networks fails to effectively deal with missing data. This paper presents two methods to account for missing data. One method recasts the incomplete data set into a complete data set and then learns Bayesian networks from the complete data set. The other learns Bayesian networks directly from the incomplete data set and this method is gradually correct. The experimental results show that the former provides accurate results, but is inefficient; while the latter is highly efficient, and can obtain good results when the data set is large. Furthermore, both methods perform better than other methods that deal with missing data.