Vulnerability- and Diversity-Aware Anonymization of Personally Identifiable Information for Improving User Privacy and Utility of Publishing Data.

Abdul Majeed,Sungchang Lee,Farman Ullah

doi:10.3390/s17051059

Abdul Majeed, Sungchang Lee + Show 1 more

Open Access

https://doi.org/10.3390/s17051059

Copy DOI

Journal: Sensors (Basel, Switzerland)	Publication Date: May 8, 2017
Citations: 25	License type: CC BY 4.0

Affiliation: Korea Aerospace University

Abstract

Personally identifiable information (PII) affects individual privacy because PII combinations may yield unique identifications in published data. User PII such as age, race, gender, and zip code contain private information that may assist an adversary in determining the user to whom such information relates. Each item of user PII reveals identity differently, and some types of PII are highly identity vulnerable. More vulnerable types of PII enable unique identification more easily, and their presence in published data increases privacy risks. Existing privacy models treat all types of PII equally from an identity revelation point of view, and they mainly focus on hiding user PII in a crowd of other users. Ignoring the identity vulnerability of each type of PII during anonymization is not an effective method of protecting user privacy in a fine-grained manner. This paper proposes a new anonymization scheme that considers the identity vulnerability of PII to effectively protect user privacy. Data generalization is performed adaptively based on the identity vulnerability of PII as well as diversity to anonymize data. This adaptive generalization effectively enables anonymous data, which protects user identity and private information disclosures while maximizing the utility of data for performing analyses and building classification models. Additionally, the proposed scheme has low computational overheads. The simulation results show the effectiveness of the scheme and verify the aforementioned claims.

Highlights

Most organizations collect relevant customer data to improve service quality
A threshold value for the comparison of each class diversity is provided, and considerable attention is paid to low diverse classes to overcome privacy breaches; and (4) we proposed an adaptive generalization scheme for anonymizing data considering the identity vulnerability of QIs as well as the diversity of SAs in equivalence classes to improve both user privacy and utility as compared to existing privacy-preserving data publishing (PPDP) algorithms
To benchmark the proposed method with other existing methods, the proposed method is compared with InfoGain Mondrian [41] and information-based anonymization for classification given k (IACk) [43], both of which have been demonstrated as being better than other methods in terms of utility and privacy when anonymizing data

Summary

Introduction

Most organizations collect relevant customer data to improve service quality. The excessive amount of collected data often contains information on customer demographics, finances, interests, activities, and medical status. Organizations are not willing to publish person-specific data because personally identifiable information (PII) often leads to unique individual identification. In PPDP, explicit identifiers (e.g., name) are removed, and QIs are generalized or suppressed to protect the SAs of individuals from disclosure. K-anonymity [18] provides sufficient protection because of its conceptual simplicity. It is a well-known PPDP model, and due to algorithmic advances in creating k-anonymous datasets [23,24,25,26], it has become a benchmark. In this example, attributes are divided into two groups: QIs and SA.

Methods

Results

Conclusion