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

Abstract

Recently, Big Data processing becomes crucial to most enterprise and government applications due to the fast growth of the collected data. However, this data often includes private personal information that arise new security and privacy concerns. Moreover, it is widely agreed that the sheer scale of big data makes many privacy preserving techniques unavailing. Therefore, in order to ensure privacy in big data, anonymization is suggested as one of the most efficient approaches. In this paper, we will provide a new detailed classification of the most used non-cryptographic anonymization techniques related to big data including generalization and randomization approaches. Besides, the paper evaluates the presented techniques through integrity, confidentiality and credibility criteria. In addition, three relevant anonymization techniques including k-anonymity, l-diversity and t-closeness are tested on an extract of a huge real data set.

Privacy Protection in Intelligent Healthcare Atomosphere: The Present and Prospect

“…A secure smart healthcare system defined by Zhang et al offers user privacy protection by providing fine-grained access control to smart healthcare cloud data. Cipher text-policy attribute-based encryption is a promising cryptographic primitive [11].…”

The Interface of Privacy and Data Security in Automated City Shuttles: The GDPR Analysis

“…Tokenization is another pseudonymization scheme that can be considered within the vehicular environment. It is the process of replacing sensitive characters by other random non-sensitive values that cannot be mathematically computed from the data source length [49]. Although, when applied to location data, even if the identities are replaced by pseudonyms, individuals can still be singled out and tracked, as demonstrated by De Montjoye et al [50].…”

“…There is a wide spectrum of anonymization techniques that can be applied to protect vehicular data, but their definitions are noticeably overlapping and commonly mistaken [49]. Data randomization (also called noising) and data generalization are the core anonymization families according to opinion 05/2014 of the WP29 [35].…”

“…Data masking is frequently considered as a further randomization technique that refers to the process of hiding true values to make the sensitive information inconceivable, such as by replacing characters with asterisks [49]. Nevertheless, data masking remains a generic term that refers to the process of hiding true values [33], which at the end can refer to pseudonymization or anonymization [56].…”

“…Such difference reduces the inference risk but does not completely eliminate it [59]. • t-closeness: is a refinement of the l-diversity theory that aims to set a t-threshold by computing the resemblance of a sensitive value distribution within the equivalent class in comparison to the attribute distribution in the whole dataset [49]. More specific to location data, obfuscation techniques are meant to purposely return incorrect location information to make the unauthorized tracking difficult [60].…”