Data privacy with heuristic anonymisation
Sevgi Arca, Rattikorn Hewett · International Journal of Information and Computer Security · 2023
Abundance of data makes privacy more vulnerable than ever as it increases the attackers' ability to infer confidential data from multiple data sources. Anonymisation protects data privacy by ensuring that critical data are non-unique to any individual so that we can conceal the individual's identity. Existing techniques aim to minimally alter the original data so that either the anonymised data or its analytical results (e.g., classification) will not disclose certain privacy. Our research aims both. This paper presents HeuristicMin, an anonymisation approach that applies generalisations to satisfy user-specified anonymity requirements while maximising data retention (for analysis purposes). Unlike others, by exploiting monotonicity property of generalisation and simple heuristics for pruning, HeuristicMin provides an efficient exhaustive search for optimal generalised data. The paper articulates different meanings of optimality in anonymisation and compares HeuristicMin with well-known approaches analytically and empirically. HeuristicMin produces competitive results on the classification obtained from the anonymised data.