Energy cost and machine learning accuracy impact of k-anonymisation and synthetic data techniques - Abstract

Aug 3, 2026

To address increasing societal concerns about privacy and climate, the EU adopted GDPR and committed to the Green Deal. Research has examined both the energy efficiency of software and the accuracy of machine learning models trained on anonymised data sets, while newer work studies how privacy-enhancing techniques affect energy consumption and predictive quality, focusing on k-anonymity. This paper analyses energy and accuracy across two phases: privacy-enhancing processing and subsequent model training, using k-anonymisation and synthetic data.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Introduction

Research questions

Energy cost and machine learning accuracy impact of k-anonymisation and synthetic data techniques Pepijn de Reus Master Artificial Intelligence University of Amsterdam Amsterdam, The Netherlands [p.dereus@uva.nl](p.dereus@uva.nl) Ana Oprescu Complex Cyber Infrastructure University of Amsterdam Amsterdam, The Netherlands [a.m.oprescu@uva.nl](a.m.oprescu@uva.nl) Koen van Elsen Institute for Informatics University of Amsterdam Amsterdam, The Netherlands [k.m.j.vanelsen@uva.nl](k.m.j.vanelsen@uva.nl) arXiv :2305 .07116v2 [ cs .LG] 29 Oct 2023 Abstract—To address increasing societal concerns regarding privacy and climate, the EU adopted the General Data Protection Regulation (GDPR) and committed to the Green Deal. Considerable research studied the energy efficiency of software and the accuracy of machine learning models trained on anonymised data sets. Recent work began exploring the impact of privacyenhancing techniques (PET) on both the energy consumption and accuracy of the machine learning models, focusing on kanonymity. As synthetic data is becoming an increasingly popular PET, this paper analyses the energy consumption and accuracy of two phases: a) applying privacy-enhancing techniques to the concerned data set, b) training the models on the concerned privacy-enhanced data set. We use two privacy-enhancing techniques: k-anonymisation (using generalisation and suppression) and synthetic data, and three machine-learning models. Each model is trained on each privacy-enhanced data set. Our results show that models trained on k-anonymised data consume less energy than models trained on the original data, with a similar performance regarding accuracy. Models trained on synthetic data have a similar energy consumption and a similar to lower accuracy compared to models trained on the original data. Index Terms—k-anonymity, synthetic data, machine learning, energy consumption of machine learning, energy consumption of artificial intelligence, privacy-enhancing machine learning I. INTRODUCTION To address climate change the European Commission has set a goal of reducing net carbon emission to zero in 2050 [1] . The amount of publications on Artificial Intelligence (AI) has increased more than fivefold over the last decade [2] . Early studies already warned of the dangers of using digitalisation without rebound considerations regarding the environment [3] and identified the need for a more fine-grained analysis of digital processes with respect to their ecological footprint [4] . Nowadays, several international initiatives, such as the EU Green Deal, aim to reduce carbon emissions. With this goal in mind, the EU aims to make data centres and ICT infrastructures climate-neutral by 2030 and aims to make use of artificial intelligence and other digital technologies to reduce the impact of climate change [5] . This has encouraged more research with a focus on energy consumption within machine learning [6]–[9] . Apart from the energy aspect there is growing concern about privacy amongst citizens of all ages [10]. To ensure the privacy of those whose data are collected the General Data Protection Regulation (GDPR) has been adopted in Europe in 2016 . The GDPR regulates that all citizens in Europe have control over their personal data [11] . There is one exception to the GDPR; it does not apply to anonymised data. In the GDPR, anonymous data is defined as: ”information which does not relate to an identified or identifiable natural person or to personal data rendered anonymous in such a manner that the data subject isnot or no longer identifiable.” [12] . Therefore it is interesting to look into methods to anonymise data such that data can be shared without GDPR constraints, especially in light of the upcoming Data Act 1. One common method for enhancing privacy of data is kanonymity, using either generalisation and suppression [13] or micro-aggregation [14] . Apart from k-anonymity, synthetic data is becoming incre

FAQs

How do synthetic data models compare to original-data models?

Models trained on synthetic data show similar energy consumption, but similar to lower accuracy compared with models trained on the original data.

Aug 3, 2026

Upload by Ava Thompson