Towards Automatic Identification Of Javascript-Oriented Machine-Based Tracking

Machine-based tracking is a type of behavior that extracts information on a user's machine, which can then be used for fingerprinting, tracking, or pro filing purposes. In this paper, we focus on JavaScript-oriented machine-based tracking as JavaScript is widely accessible in all browsers. We find that coarse features related to JavaScript access, cookie access, and URL length / subdomain information can perform well in a supervised machine learning classifier that can identify machine-based trackers with 97.7% accuracy. We then use the classifier on real-world datasets based on 30-minute web-site crawls of different types of websites - including websites that target children and websites that target a popular audience - and find 85%+ of all websites utilize machine-based tracking, even when they target a regulated group (children) as their primary audience.