Development of an association rule hiding algorithm for privacy preserving in market basket databases

Abstract

Association Rule Hiding is achieved by applying privacy preserving data mining techniques on a database. It becomes necessary before revealing the database to the third party. Major limitations of popular association rule hiding algorithms are failing into hiding all sensitive association rules, losses in terms of large number of non sensitive association rules, generation of ghost rules and false rules during the process of association rule hiding. These drawbacks have a greater impact the factors like privacy, correctness and usefulness of the sanitized database. Trustworthiness of inferences, conclusions and results extracted from sanitized database is affected. In this paper, an efficient algorithm named as selective flip bit was proposed as a solution for association rule hiding that hides all sensitive association rules by generating very less number of lost rules and zero generation of ghost rules and false rules. Developed algorithm was tested on both artifactual and real life databases. A software tool was developed to implement the selective flip bit algorithm on real life database. Another tool was also developed for the performance evaluation of developed rule hiding algorithm. Results indicate that the proposed selective flip bit algorithm is highly efficient in terms of hiding sensitive association rules along with retaining maximum non sensitive association rules as compared to the algorithms in the same field.