PRFP: Privacy-respecting and Accurate Device Fingerprint Identification

Abstract

With the rapid development of the mobile device, various online transactions have attracted a large number of black markets. Malicious users forge mobile equipments to register accounts, receive coupons, and get the best deal, which cause serious losses to merchants, enterprises, and other consumers. In order to combat these threats, device fingerprinting technology has been studied to derive unique identification and defeat malicious registrations and coupons from the same device. However, existing device fingerprinting methods require collecting sensitive information from users' devices and hence leak privacy. Although there are studies protecting privacy by adding noises or perturbations to collected device information, they significantly reduce the usability of collected sensor data and hence degrade the accuracy of device fingerprinting. In this paper, we design a framework named PRFP, which achieves high accuracy of fingerprinting devices while effectively protecting users' privacy. It enables deep learning models to extract informative intermediate feature representations related to device fingerprinting from raw device, while automatically removing user-related private information as much as possible. Our experimental results with a large device fingerprinting dataset show that PRFP can effectively protect users' privacy while only introducing less than 4% accuracy loss compared to fingerprinting devices with raw data.