Privacy by Projection: Federated Population Density Estimation by Projecting on Random Features.

Area of Science:

• Computer Science
• Data Science
• Mobile Computing

Background:

• Population density estimation commonly uses Kernel Density Estimation (KDE) with centralized location data.
• Centralized data collection raises significant user privacy concerns.
• Existing privacy-preserving methods may compromise estimation accuracy.

Purpose of the Study:

• To propose a privacy-preserving Federated Kernel Density Estimation (KDE) framework for population density estimation.
• To ensure user location data remains on devices while providing privacy guarantees against malicious servers.
• To achieve a superior trade-off between estimation utility and user privacy.

Main Methods:

• Developed a Federated Random Fourier Feature (RFF) KDE approach.
• Utilized random feature representation to project user location data onto spatially delocalized basis functions.
• Ensured irreversible projection for enhanced privacy, preventing precise user localization.

Main Results:

• Federated RFF KDE demonstrated a superior utility-privacy trade-off compared to state-of-the-art methods like GeoInd.
• Adjusting the number of basis functions per user further optimized the privacy-utility balance.
• Analytical bounds on localization were derived based on areal unit size and kernel bandwidth.

Conclusions:

• Federated RFF KDE offers an effective and privacy-preserving solution for population density estimation using crowdsourced mobile location data.
• The method successfully balances the need for accurate density estimation with robust user privacy.
• This framework represents a significant advancement in secure location-based data analysis.