Extending the Applicability of Association Rules

An important issue that needs to be addressed when using association rules is the validity of the rules for new situations. Rules are typically derived from the patterns in a particular dataset. When the conditions under which the dataset has been obtained change, a new situation is said to have risen. Since the conditions existing at the time of observation could aaect the observed data, a change in those conditions could imply a changed set of rules for a new situation. Using the set of rules derived from the dataset for an earlier situation could lead to wrong decisions. In this paper, we provide a model explaining the diierence between the sets of rules for diierent situations. Our model is based on the concept of rule-generating groups that we call caucuses. Using this model, we provide a simple technique, called Linear Combinations, to get a good estimate of the set of rules for a new situation. Our approach is independent of the core mining process, and so can be easily implemented with any speciic technique for association rule mining. In our experiments using controlled datasets, we found that we could get up to 98.3% accuracy with our techniques as opposed to 26.6% when directly using the results of the old situation.