Mining Association Rules With Constraints

Association rule mining is one of the most important tasks of Data Mining, which is to find the rules with the support and confidence greater than the thresholds.There are many disadvantages in the traditional association rule mining process. For example, users can't express their focus and control the mining process, and also it's not efficient. But all these disadvantages can be resolved in the interactive association rule mining process. There are two ways to mine all rules which satisfy constraints the user defined. One is Generate and Test, in which all rules including large number of useless rules are generated firstly, then tested with constraints. The other is to push constraints into the frequent itemsets mining process so that the results are just those itemsets satisfying constraints. Because the results become smaller, the latter way is more efficient. According to their properties, constraints can be devided into anti-monotone, monotone, succinct, convertible and inconvertible constraints. All the anti-monotone, monotone and succinct constraints can be pushed deeply into the Apriori algorithm, while convertible constraints can only be pushed into frequent-pattern growth algorithm and inconvertible constraints can be done in neither. The CAP+ fram can implement different kinds of constraints. In contrast with CAP, monotone, convertible, inconvertible constraints are all considered. According to the property of monotone constraint, the new algorithm named MCA, which is based on the new notion of based monotone sets, combined with anti-monotone constraint, can simultaneously implement monotone constraint and anti-monotone constraints with good efficiency.