Energy cost and accuracy impact of k-anonymity

European Union has aggregated the current societal concerns into two seemingly orthogonal directions: the Green Deal and the GDPR. In this paper, we begin to analyse trade-offs in preserving privacy, learning from the data, and saving energy. Considerable research studied the energy efficiency of software and the accuracy of machine learning models trained on anonymised datasets. However, to the best of our knowledge, no research has been conducted on the impact of anonymisation techniques on energy consumption. We measure the impact of anonymisation on the energy consumption and on the accuracy of machine learning models.We find that the k-value has a statistically significant impact on the energy consumption of the chosen anonymization algorithms. In terms of the accuracy of machine learning models, the generalization and suppression performs better in almost all cases, provided that proper anonymization hierarchies are used in the anonymization process. However, we find that for the larger and more complex dataset, the reduction in accuracy is limited while there is a significant difference in energy consumption. Thus when considering energy consumption we conclude that for larger datasets it might be worthwhile to consider using microaggregation over generalization and suppression.

I. Introduction

Modern society is digitalizing. The internet has transformed the way we receive information, conduct business and present ourselves. In this era of digitalization, more and more business models are based on the “new gold” that consists of user and process data [32] . The European Union (EU) uses the General Data Protection Regulation (GDPR) [45] to protect their citizens and to give businesses a regulatory framework on which they could construct their business models. Interestingly, if data is sufficiently anonymized then it can be used for any purpose, be collected without consent and be stored for an indefinite amount of time [37] .