Browser Fingerprinting

Abstract

Tracking Internet users have several purposes; for example preventing online fraud or measuring and analysing online advertisements. The most common current methods for tracking users involve storing unique identifiers locally on the user's device, a method which may be restricted by law in the future. This thesis examines the possibility to replace suchmethods with the method of browser fingerprinting. A browser fingerprint is a composition of information gathered from a web browser. Using data from several sources, we have analysed which features to extract to create a unique fingerprint. Within the scope of machine learning, we have designed online algorithms which can be used for telling Internet users apart by their browser fingerprints. The results from these algorithm show that if the data is preprocessed and partitioned adequately, users can be identified with high accuracy over time periods spanning over several days, even weeks in some cases. Based on our analysis, we can conclude that machine learning is a promising approach for solving the problem of telling Internet users apart and that extremely naive solutions are sufficient, in certain applications.