The Compromise of Data Privacy in Predictive Performance

Privacy-preservation has become an essential concern in many data mining applications since the emergence of legal obligations to protect personal data. Thus, the notion of Privacy-Preserving Data Mining emerged to allow the extraction of knowledge from data without violating the privacy of individuals. Several transformation techniques have been proposed to protect the privacy of individuals. However, their application does not guarantee a null risk of an individual being reidentified. Furthermore, and most importantly, for this paper, the application of such techniques may have a considerable impact on the utility of data and their use in predictive and descriptive tasks. In this paper, we present a study to provide key insights concerning the impact of privacy-preserving techniques in predictive performance. Unlike previous work, our main conclusions point towards a noticeable impact of privacy-preservation techniques in predictive performance.