Learning maximum likelihood semi-naive Bayesian network classifier

We propose a technique to construct a sub-optimal semi-naive Bayesian network when given a bound on the maximum number of variables that can be combined into a node. We theoretically show that our approach has less computation cost when compared with the traditional semi-naive Bayesian network. At the same time, we can obtain a resulting sub-optimal structure according to the maximum likelihood criterion. We conduct a series of experiments to evaluate our approach. The results show our approach is encouraging and promising.