Privacy Assessment Methodology for Machine Learning Models and Data Sources
Rudolf Erdei, Emil Pasca, Daniela Delinschi, Anca Avram, Ionela Chereja, Oliviu Matei

Abstract. The widespread use of machine learning amplifies privacy risks both at the level of training data and at the level of the resulting models. This paper proposes a methodology for the joint privacy assessment of machine learning models and the data sources used to build them. The approach combines a structured inventory of data sources, threat scenarios and privacy-relevant model properties (memorisation, leakage potential, re-identification risk) with quantitative indicators that can be computed during the model lifecycle. The methodology supports compliance with privacy regulations and enables informed trade-offs between utility and privacy, with case studies drawn from agricultural and IoT data domains.

Keywords: privacy assessment; machine learning; data sources; privacy risk; data protection

Rudolf Erdei, Emil Pasca, Daniela Delinschi, Anca Avram, Ionela Chereja, Oliviu Matei, "Privacy Assessment Methodology for Machine Learning Models and Data Sources", Proc. 19th SOCO Int. Conf. on Soft Computing Models in Industrial and Environmental Applications, Springer, 2024, 2023.

Reference: Proc. 19th SOCO Int. Conf. on Soft Computing Models in Industrial and Environmental Applications, Springer, 2024.

Other publications