A standardised differential privacy framework for epidemiological modeling with mobile phone data.

Area of Science:

• Epidemiology
• Data Science
• Public Health

Background:

• Mobile phone data are crucial for tracking human mobility during pandemics like COVID-19.
• Protecting individual privacy in location data is essential for ethical public policy and epidemiological modeling.
• Traditional data aggregation methods fail to guarantee anonymity, necessitating advanced privacy-preserving techniques.

Purpose of the Study:

• To evaluate differential privacy as an anonymization tool for mobile phone mobility data in epidemiological models.
• To assess the impact of statistically quantified noise on common epidemiological metrics.
• To determine the robustness of differential privacy for public health applications.

Main Methods:

• Applied differential privacy by adding statistical noise to mobile phone location data.
• Analyzed the bias introduced to ten common epidemiological metrics.
• Quantified the relationship between noise levels and metric accuracy using a count transition matrix.
• Developed a modular software pipeline for reproducibility.

Main Results:

• Many epidemiological metrics remained accurate, close to non-private values, with noise levels below 20 (ϵ = 0.05 per release).
• Differential privacy demonstrated a statistically verifiable protection against identifiability.
• The study confirmed the utility of privacy-preserving mobility data for public health.

Conclusions:

• Differential privacy offers a robust solution for anonymizing mobility data while retaining valuable epidemiological insights.
• This approach balances individual privacy with the need for population-level data in public health.
• The developed framework supports further research and application of privacy-preserving epidemiological modeling.