Abstract
This article provides a state-of-the-art summary of location privacy issues and geoprivacy-preserving methods in public health interventions and health research involving disaggregate geographic data about individuals. Synthetic data generation (from real data using machine learning) is discussed in detail as a promising privacy-preserving approach. To fully achieve their goals, privacy-preserving methods should form part of a wider comprehensive socio-technical framework for the appropriate disclosure, use and dissemination of data containing personal identifiable information. Select highlights are also presented from a related December 2021 AAG (American Association of Geographers) webinar that explored ethical and other issues surrounding the use of geospatial data to address public health issues during challenging crises, such as the COVID-19 pandemic.
Introduction
The management of public health crises frequently necessitates high-resolution geographic data to identify clusters and interrupt transmission chains. However, this requirement creates a fundamental tension with the right to individual privacy. As location-tracking technologies—from smartphones to wearables—become ubiquitous, the risk of "geoprivacy" violations increases. This paper explores contemporary methods to reconcile the "common good" of public health with individual privacy protections, focusing on Synthetic Data Generation (SDG) and socio-technical frameworks.
The Evolution of the Geoprivacy Conflict
The utility of micro-scale geographic analysis is well-established, famously exemplified by John Snow’s 1854 cholera map. Snow’s success relied on knowing the precise locations of individual cases. In the modern era, however, traditional de-identification techniques, such as coordinate blurring or masking, are increasingly vulnerable to re-identification attacks. The core challenge lies in sharing high-quality spatial data without allowing an individual’s home or workplace to be reverse-engineered from the dataset.
Emerging Privacy-Preserving Methodologies
Several state-of-the-art approaches have emerged to address these vulnerabilities:
Synthetic Data Generation (SDG)
SDG utilizes machine learning models—such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), or Bayesian networks—trained on real datasets. These models generate entirely new, "synthetic" records that maintain the statistical properties and multi-variable relationships of the original data without a one-to-one mapping to real individuals. SDG often provides higher data utility than traditional masking because it preserves complex spatial patterns.
Decentralized Proximity Sensing
A significant advancement during the COVID-19 pandemic was the shift from GPS-based tracking to Bluetooth-based proximity sensing, notably the Apple-Google Exposure Notification API. By using decentralized Bluetooth "keys" rather than absolute coordinates, these systems decouple "who you were near" from "where you were," effectively preserving geoprivacy while achieving public health goals.
Software-Only Access and Differential Privacy
Other methods include "software-only" access models, where researchers send analysis code to a secured environment rather than receiving raw data. Additionally, Differential Privacy provides a mathematical framework for adding controlled "noise" to datasets, though it often involves a significant trade-off between privacy levels and data utility.
Socio-Technical and Cultural Considerations
Privacy is not a universal constant; it is shaped by cultural and social contexts. Research indicates that "collectivist" cultures (e.g., South Korea) often show higher acceptance of location-aware pandemic measures compared to "individualist" cultures, driven by a different perception of social benefit versus personal risk. Furthermore, a survey of 32 jurisdictions revealed that 75%—primarily in Europe—opted against location tracking for contact tracing due to geoprivacy concerns, illustrating the profound impact of policy on health strategy.
Limitations and Future Research
Despite the promise of SDG, several hurdles remain:
- Longitudinal Complexity: Creating synthetic versions of "trajectories" (movement over time) is significantly more complex than synthesizing tabular data.
- Geographic Plausibility: Machine learning models can sometimes generate "implausible points," such as coordinates located in the ocean or on inaccessible terrain. Future models must integrate geographic constraints like road networks.
- The "Rare Event" Problem: Generative models excel at capturing averages but struggle with outliers. In public health, where the first case of a new variant is a critical outlier, this limitation is significant.
Conclusion
Technical solutions alone are insufficient to protect geoprivacy. A comprehensive socio-technical framework is required, incorporating legal safeguards such as "sunset clauses" to ensure that data collection and retention end once a crisis concludes. By combining advanced methods like SDG with robust policy frameworks, society can better navigate the balance between public health necessity and individual privacy rights.
Summary updated