A (k, p)-anonymity Framework to Sanitize Transactional Database with Personalized Sensitivity

Abstract

In recent years, analyzing transactional data has become an important data analytic task since it can discover important information in several domains, for recommendation, prediction, and personalization. Nonetheless, transactional data sometimes contains sensitive and confidential information such as personal identifiers, information aboutsexual orientations, medical diseases, and religious beliefs. Such information can be analyzed using various data mining algorithms, which may cause security threats to individuals. Several algorithms were proposed to hide sensitive information in databases but most of them assume that sensitive information is the same for all users, which is an unrealistic assumption. Hence, this paper presents a (k, p)-anonymity framework to hide personal sensitive information. The developed ANonymity for Transactional database (ANT) algorithm can hide multiple pieces of sensitive information in transactions. Besides, it let users assign sensitivity values to indicate how sensitive each piece of information is. The designed anonymity algorithm ensures that the percentage of anonymized data does not exceed a predefined maximum sensitivity threshold. Results of several experiments indicate that the proposed algorithm outperforms thestate-of-the-art PTA and Gray-TSP algorithms in terms of information loss and runtime.