Discovering reliable evidence of data misuse by exploiting rule redundancy

Abstract

Big Data offers opportunities for in-depth data analytics and advanced personalized services. Yet, while valuable, data analytics might rely on data that should not have been used due to, e.g., privacy constraints from the data subject or regulations. As decision makers and data controllers often act outside any control mechanism and with no requirement of transparency, it is challenging to verify whether constraints on data usage are actually satisfied. In this work, we relate the problem of finding evidence of data misuse to the identification of unique decision rules, i.e. rules that have likely been used for decision making. Accordingly, we propose an approach to find reliable evidence of data misuse in the context of classification problems using association rule mining, along with novel metrics to assess the level of redundancy among decision rules. Our proposed approach is able to identify the use of sensitive information in decisional processes along with their context. We evaluated our approach through both controlled experiments and two case studies using real-life event data. The results show that our approach finds more reliable evidence of data misuse compared to previous work.