Prediction Mining – An Approach to Mining Association Rules for Prediction

Abstract

An interesting application of association mining in the context temporal databases is that of prediction. Prediction is to use the antecedent of a rule to predict the consequent of the rule. But not all of association rules may be suitable for prediction. In this paper, we investigate the properties of rules for prediction, and develop an approach called prediction mining — mining a set of association rules that are useful for prediction. Prediction mining discovers a set of prediction rules that have three properties. First, there must be a time lag between antecedent and consequent of the rule. Second, antecedent of a prediction rule is the minimum condition that implies the consequent. Third, a prediction rule must have relatively stable confidence with respect to the time frame determined by application domain. We develop a prediction mining algorithm for discovering the set of prediction rules. The efficiency and effectiveness of our approach is validated by experiments on both synthetic and real-life databases, we show that the prediction mining approach efficiently discovers a set of rules that are proper for prediction.

This research was supported in part by NSF Digital Government Grant No. EIA-0091530, USDA RMA Grant NO. 02IE08310228, and NSF EPSCOR, Grant No. EPS-0091900.