Vasil Zlatanov is a Developer at Netcraft, where he works on the visual classification of phishing websites. He graduated from Imperial College London in 2019 with a MEng in Electrical & Electronic Engineering.
Defeating Phishing with Convolutional Networks
Netcraft operates an industry-leading feed of validated phishing websites, which is licensed by all leading web browsers to protect their users. With the development of the web, phishing website have become ever more complicated and difficult to detect in a perpetual cat and mouse game. Even though phish try to obfuscate their source as much as possible, they will not significantly deviate from the appearance of their target. Using Deep Convolutional Neural Networks on fully rendered phishing websites allows human-like classification of websites based on how they look. Initially developed as final year project, convolutional networks are currently used in production in Netcraft in collaboration with previous final year projects. We demonstrate the ability of networks to learn features such as logos, layouts, colours and elements and their effectiveness at the visual classification phishing websites. We look at how visual classifications is used at Netcraft, while also outlining the challenges involved with using Machine Learning in production.