Privacy Preserving Data Analysis with the Encode, Shuffle, Analyse Architecture in Vehicular Data Sharing

In recent years, vehicles have become smarter, with more data being collected and analyzed. With further digitalisation, the importance of data within and around vehicles is only going to increase, allowing stakeholders to generate new and more insights from that data. With this, the threat of privacy invasion and the de-identification of vehicles and vehicle users becomes more pressing. In this work, we implement a prototype of a variant of distributed differential privacy (DP), called Encode, Shuffle, and Analyse (ESA) architecture, on a real-world vehicular dataset for vehicle energy consumption analysis. An analysis for energy consumption standard statistics and a basic neural network prediction model are set up to elaborate the utility, privacy trade-off. The results are then compared to the same analysis in a non-private setting. This work contributes to the basis of knowledge by providing a prototype of the ESA architecture in vehicle data analysis for energy demand prediction in urban areas. The results identify important parameters to balance privacy and utility and provides actionable insights for regulators and vehicle manufacturers to preserve vehicle privacy. Future work should further investigate possible attack scenarios to achieve a higher understanding of the privacy guarantees of this architecture.