Improving the K2 Algorithm Using Association Rule Parameters

A Bayesian network is an appropriate tool to work with the uncertainty that is typical of real-life applications. Bayesian network arcs represent statistical dependence between different variables and can be automatically elicited from database by Bayesian network learning algorithms such as K2. In the data mining field, association rules can also be interpreted as expressing statistical dependence relations. In this paper we present an extension of K2 called K2-rules that exploits a parameter normally defined in relation to association rules for learning Bayesian networks. We compare K2-rules with K2 and TPDA on the problems of learning four Bayesian networks. The experiments show that K2-rules improves both K2 and TPDA with respect to the quality of the learned network and K2 with respect to the execution time.