Combination of Access Control and De-Identification for Privacy Preserving in Big Data

Abstract

Despite of its emergence and advantages in various domains, big data still suffers from major disadvantages. Timeless, scalability, and privacy are the main problems that hinder the advance of big data. Privacy preserving has become a wide search era within the scientific community. This paper covers the problem of privacy preserving over big data by combining both access control and data de-identification techniques in order to provide a powerful system. The aim of this system is to carry on all big data properties (volume, variety, velocity, veracity, and value) to ensure protection of users' identities. After many experiments and tests, our system shows high efficiency on detecting and hiding personal information while maintaining the utility of useful data. The remainder of this report is addressed in the presentation of some known works over a privacy preserving domain, the introduction of some basic concepts that are used to build our approach, the presentation of our system, and finally the display and discussion of the main results of our experiments.

A Classification of non-Cryptographic Anonymization Techniques Ensuring Privacy in Big Data

“…Approaches Techniques Integrity Confidentiality Credibility Generalization K-anonymity based on generalization [22], [23], [24], [25] No No Yes L-diversity [2], [14], [23], [25] Yes No Yes T-closeness [2], [14], [26] Yes Yes Yes Randomization K-anonymity based on suppression [20], [24], [27] No No Yes…”

“…In addition, it is impossible to implement the inference attacks against an 'l-diverse' data set with certitude of 100% [14]. In the literature, there exist three models of l-diversity which are Distinct, Entropy and Recursive models [24], [26]. However, the distinct l-diversity technique is the most used where each bucket in the data set contains only distinct values.…”

“…When the distance between the distribution of a sensitive attribute in the equivalence class and the distribution of the attribute in the whole table is less than a threshold t, then the equivalence class is called "have t-closeness" [2], [26]. As said before, l-diversity technique is not able to resist against several attacks.…”