Finding minimum confidence threshold to avoid derived rules in association rule mining

Abstract

Data in data warehouse often contains sensitive information, the concept of PrivacyPreserving has recently been proposed in response to the concerns of preservingsensitive information derived from published rules. A number of privacy preservingdata publishing (PPDP) have been proposed. In this paper an algorithm proposed forhiding published rules that leads to disclosure of sensitive information by determiningthe confidence value of those rules from the raw data before running association rulemining using prior and posterior probabilities of generated rules and pass thoseconfidence values to data miner to take it in his account when determining minimumconfidence threshold in association rule mining algorithms .The experimental resultsshow that the run time for deriving sensitive rules is stabile for different confidencevalues in comparison with other methods running linear programming methods forfinding sensitive published rules. The most derived rules from goal rules (the rulesderived from sensitive rules with minimum confidence value) located between 0.5and 0.8 and these range of confidence values are critical values for data miner, finallyexperimental results shows that with support values %40,%58, and %63 still there isamount of derived published rules appears, and these results means that even withlarge minimum support threshold still derived published rules appears in associationrule algorithms.