Data Mining and Privacy

Data Mining and Privacy

Esma Aïmeur (Université de Montréal, Canada) and Sébastien Gambs (Université de Montréal, Canada)
Copyright: © 2012 |Pages: 8
DOI: 10.4018/978-1-61350-323-2.ch110
OnDemand PDF Download:
$30.00
List Price: $37.50

Chapter Preview

Top

Background

The area of privacy-preserving data mining can still be considered in its infancy but there are already several workshops (usually held in collaboration with different data mining and machine learning conferences), two different surveys (Verykios et al., 2004; Výborný, 2006) and a short book (Vaidya, Clifton & Zhu, 2006) on the subject. The notion of privacy itself is difficult to formalize and quantify, and it can take different flavours depending on the context. The three following scenarios illustrate how privacy issues can appear in different data mining contexts.

  • Scenario 1: A famous Internet-access provider wants to release the log data of some of its customers (which include their personal queries over the last few months) to provide a public benchmark available to the web mining community. How can the company anonymize the database in such a way that it can guarantee to its clients that no important and sensible information can be mined about them?

  • Scenario 2: Different governmental agencies (for instance the Revenue Agency, the Immigration Office and the Ministry of Justice) want to compute and release some joint statistics on the entire population but they are constrained by the law not to communicate any individual information on citizens, even to other governmental agencies. How can the agencies compute statistics that are sufficiently accurate while at the same time, safeguarding the privacy of individual citizens?

  • Scenario 3: Consider two bioinformatics companies: Alice Corporation and Bob Trust. Each company possesses a huge database of bioinformatics data gathered from experiments performed in their respective labs. Both companies are willing to cooperate in order to achieve a learning task of mutual interest such as a clustering algorithm or the derivation of association rules, nonetheless they do not wish to exchange their whole databases because of obvious privacy concerns. How can they achieve this goal without disclosing any unnecessary information?

Complete Chapter List

Search this Book:
Reset