An efficient node ordering method using the conditional frequency for the K2 algorithm

• A child has a better conditional frequency (or probability) under a correct parent than under an incorrect one.
• The proposed method infers a correct causal relation by evaluating the conditional frequency through our scoring function.
• Experimental results indicate that our proposed method outperforms previous methods in the accuracy and the consumed times.

In Bayesian networks, the K2 algorithm is one of the most effective structure-learning methods. However, because the performance of the K2 algorithm depends on node ordering, more effective node ordering inference methods are needed. In this paper, we therefore introduce a new node ordering algorithm based on a novel scoring function. Because a child has a better conditional frequency or probability under a correct parent than an incorrect one, we have designed a novel scoring function to evaluate this conditional frequency. Given two variables, our scoring function infers which is the better parent variable. Consequently, the proposed method infers candidate parents by considering all pairs of variables; it then uses these parents as input for the K2 algorithm. Experimental results indicate that our proposed method outperforms previous methods.