Leveraging Gaussian Noise for Balancing Differential Privacy Guarantees and Utility in Mobility Reports

Smart cities optimize public transportation systems by analyzing mobility data, but due to its highly sensitive nature, they must find a balance between its utility and the protection of privacy. For safeguarding personal information differential privacy has emerged as the gold standard. However, repeated data use for different kinds of analysis divides the privacy budget, results in high levels of noise and significantly impacts utility. These limitations also affect the Python package dp_mobility_report which provides differentially private mobility reports using the Laplace noise mechanism. Compared to Laplace noise, the Gaussian noise mechanism has a more favorable privacy budget composition for performing several analyses on a single database. Therefore, we integrate the Gaussian noise mechanism into dp_mobility_report, and evaluate it on three datasets: Berlin (synthetic traffic simulation), Madrid (mobility survey), and Geolife (GPS trajectories). Our findings indicate that the Gaussian mechanism indeed improves utility, making it a promising alternative for producing mobility reports. However, this comes with trade-offs: the weaker privacy guarantee of (ϵ, δ)-differential privacy and increased implementation complexity.