On Estimating Platforms of Web User with JavaScript Math Object

Browser fingerprinting is a technique to identify a device using a combination of information that a server can gather from a browser. In general, the user-agent is known to be one of the most useful features for identification via browser fingerprinting. However, users can easily change the user-agent. This may lead to a decrease in the device identification accuracy. In this paper, we conducted two experiments. First, we proposed a method to estimate the platforms without using the user-agent. In particular, we used the fact that the computational result of a JavaScript math object varies depending on the platform. Using this method, we could classify 14 platforms into nine groups. Five of these uniquely identify the OS and the browser, two uniquely identify the OS and one uniquely identifies the browser. Second, we compared the accuracy of the browser fingerprint with user-agent (FP-A) and the browser fingerprint with our proposed method (FP-B). As a result, the identification accuracy rate with FP-B was only 0.4% lower than that with FP-A.