ML-CB: Machine Learning Canvas Block

Abstract With the aim of increasing online privacy, we present a novel, machine-learning based approach to blocking one three main ways website visitors are tracked online—canvas fingerprinting. Because act canvas fingerprinting uses, at its core, JavaScript program, and because many these programs reused across web, able fit several machine learning models around semantic representation potentially offending achieving accurate robust classifiers. Our supervised is trained on dataset created by scraping roughly half million websites using custom Google Chrome extension storing information related canvas. Classification leverages our key insight that images drawn have facially distinct appearance, allowing us manually classify files drawn; take this step further train classifiers not malleable themselves, but more-difficult-to-change, underlying source code generating images. As result, ML-CB allows for more tracker blocking.