Post mining of non-redundant association rules for sensor data estimation

Abstract

Association mining can produce many association rules. It is widely recognized that the set of association rules can rapidly grow to be unwieldy, especially when the support requirements are relatively low, making it difficult for end users to identify those that are of particular interest to them. Therefore, it is important to remove insignificant rules and prune redundant information as well as utilize the discovered information for meaningful purposes in different applications. Many researchers have considered various kinds of solutions to the above problem. For example, to efficiently mine association rules based on frequent itemsets; to mine interesting association rules based on user-specified constraints; to mine non-redundant association rules. However, the number of produced association rules is still in a large amount and not all of them are useful for the end users' requests. In this research we focus on post mining of non-redundant and informative association rules that match the user interests. The generated association rules are then applied to sensor network databases of a traffic monitoring site for missing data estimation purpose, in which data missing by a sensor is estimated using the data generated by its related sensors.