Anonymization of Daily Activity Data by Using ℓ-diversity Privacy Model

Abstract

In the age of IoT, collection of activity data has become ubiquitous. Publishing activity data can be quite useful for various purposes such as estimating the level of assistance required by older adults and facilitating early diagnosis and treatment of certain diseases. However, publishing activity data comes with privacy risks: Each dimension, i.e., the activity of a person at any given point in time can be used to identify a person as well as to reveal sensitive information about the person such as not being at home at that time. Unfortunately, conventional anonymization methods have shortcomings when it comes to anonymizing activity data. Activity datasets considered for publication are often flat with many dimensions but typically not many rows, which makes the existing anonymization techniques either inapplicable due to very few rows, or else either inefficient or ineffective in preserving utility. This article proposes novel multi-level clustering-based approaches using a non-metric weighted distance measure that enforce ℓ-diversity model. Experimental results show that the proposed methods preserve data utility and are orders more efficient than the existing methods.