Mobile Device Fingerprinting Recognition using Insensitive Information

Device fingerprinting concept is inspired from the forensic value of human fingerprints. It consists of information collected about the software and hardware of a remote computing device for the purpose of identification. This technique can be applied to areas, such as advertisement targeting, user tracking, eCommerce fraud prevention and etc. Sensitive information such as MAC address, hardware serial numbers or phone numbers are unique identifiers, however, they may leak user privacy. Therefore, device fingerprinting based on insensitive information from web browsers or mobile Apps is more favorable. In this work, we propose a passive device fingerprinting approach using a variety of zero permission features. First, a data cleansing process is set up to remove corrupted or inaccurate records. Then entropy-based feature selection method is proposed. Two models are established to create device finger prints. These models are evaluated using two datasets. The experimental results show the effectiveness of our proposed methods. The optimal model can reach average macro-F1 of 0.978 and micro-F1 of 0.982 respectively.