On the feasibility of user de-anonymization from shared mobile sensor data

Abstract

Underpinning many recent advances in sensing applications (e.g., mHealth) is the ability to safely collect and share mobile sensor data. Research has shown that even from seemingly harmless sensors (e.g., accelerometers, gyroscopes, or magnetometers) an ever expanding set of potentially sensitive user behavior can be inferred. Providing robust anonymity assurances is a principal mechanism for protecting users when data is shared (e.g., with medical professionals or friends). In this paper, we study the feasibility of user de-anonymization from mobile sensor datasets routinely collected on commodity devices (e.g., smartphones). We perform a systematic investigation to quantify the threat of de-anonymization using existing sparsity-based techniques adapted to exploit mobile sensor data characteristics. This preliminary study indicates significant threats to user anonymity exist within shared mobile sensor data and further investigation is warranted.

Unexpected Inferences from Sensor Data: A Hidden Privacy Threat in the Internet of Things

“…Such informationeven at coarse resolutioncan be used in conjunction with employment directories, white pages, tax records, and other public or private datasets to uniquely identify a tracked person [22]. Social networks have also been proven to be excellent sources of auxiliary data for the de-anonymization of smartphone sensor data [36]. Some researchers even argue that no method exists to reliably prevent the deanonymization of location data [46].…”

“…information brokers, public datasets) but can also be collected from within a user's own ecosystem of connected devices. As shown in [36], mobile sensor data can become personally identifiable through linkage with less protected data streams of the same user. Seemingly innocuous sensors could thus be used as gateways to de-anonymize sensitive streams of user activity.…”

A Literature Survey and Classifications on Data Deanonymisation

“…Sparse data share comparatively few relationships. Lane et al [49] used activity stream of the target within anonymised sets of streams gathered from other users. They designed a two-stage deanonymisation framework: the first stage follows activity relationship mining, which consist of rules that link various aspects of two captured activity types and can help to decide if two activities were probably achieved by the same individual or not.…”

Integrating Smartphones into the SmartSantander Infrastructure

“…5 Nicholas Lane and his colleagues discuss anonymity for mobile sensor datasets, 6 and Delphine Christin and her colleagues 7 extensively discuss the privacy concepts associated with mobile participatory sensing, 8,9 as well as applications for monitoring noise levels in urban areas and in-house air quality.…”