A clustering approach to anonymize locations during dataset de-identification

Abstract

Companies increasingly rely on massive amounts of data for strategic decision making purposes. In order to optimize business intelligence, companies often try to enrich their models with datasets acquired from third parties. Datasets containing sensitive attributes must be anonymized before release. For large datasets containing microdata, an often applied anonymization technique is data generalization with the goal of achieving privacy metrics such as k-anonymity. Location is an often recurring yet strategic attribute in many use cases. Multiple strategies can be employed to obfuscate precise coordinates. For example, the most significant digits can be dropped or their value can be replaced by a ZIP code. While these methods might be useful in some applications, these approaches often result in too much information loss, undermining strategic decision making. This paper proposes a novel approach to anonymize location by means of clustering. Its feasibility is evaluated and compared to traditional techniques.