Finding Causal Patterns from Frequent Itemsets

Mining association rules in transaction databases has received much attention in the field of data mining. Although progress has been made on techniques of mining association rules, the results often only indicate the mutual correlative relationships among the frequent items, paying no attention to the directional, or causal relations. For example, when a data set indicates an association between items A and B, it is often not clear whether the access of A caused the access of B, or the converse. In real world applications, however, knowing such causal relations is extremely useful for decision support. People would not only be interested in the facts that A and B are related, but also in the possible sequences and directions among the items. Mining transaction databases for this kind of knowledge offers the potential for deep analysis of business situations and finding strategies of operation. In this paper, we employ a Bayesian approach to mining causal relations from frequent itemsets. The results of our research include two algorithms based on Bayesian statistics model: a serial and diverging connection discovery algorithm (SDCD) and a converging connection discovery algorithm (CCD). Experimental results indicate that the performance of the algorithms is scalable.