文摘
Bayesian networks have emerged in recent years as a powerful data mining technique for handling uncertainty in Artificial Intelligence community. However, researchers in the classification area were not interested in Bayesian networks until the simplest kind of Bayesian networks, Naive Bayes Classifiers (NBC), came forth. From that time on, their success led to a recent furry of algorithms for learning Bayesian networks from raw data and triggered experts to explore more deeply into Bayesian networks as classifiers. Although many of learners produce good results on some benchmark data sets, there are still several problems: nodes ordering requirement, computational complexity, lack of publicly available learning tools. Therefore, this paper puts up a new method, Bayesian networks with hidden nodes, which adds some hidden nodes between correlated feature variables to Bayesian networks based on the maximal covariance criterion. Experimental results demonstrate that the proposed method is efficient and effective, and outperforms NBC and Bayesian Network Augmented Naive Bayes (BAN).