Machine Learning-based Optimal Data Trading Mechanism with Randomized Privacy Protection Scheme

In Internet of Things (IoT) systems, the vast amounts of personal data generated by IoT devices offer significant opportunities for enhancing personalized services but also introduce substantial risks of privacy leakage. Taking into account personalized privacy concerns of users, in this work we design an machine learning-based optimal data transaction mechanism that incentivizes users to share their data based on their truthful privacy valuation. We treat data with varying privacy protection levels (PPLs) as distinct sale items. By offering multiple privacy options for users to choose from, the mechanism improves data utility and encourages users to select options that reflect their true privacy valuations. By integrating machine learning into the option design in mechanism, our approach achieves optimal data utility without prior knowledge of the privacy valuation distribution. Experimental results demonstrate that even with an unknown distribution of user privacy valuations, the proposed mechanism enables users to select the optimal PPL according to their preferences, while data buyers can maximize the utility of procured data.