Email Reading Behavior-Informed Machine Learning Model to Predict Phishing Susceptibility

As phishing threats intensify, incidents like the “COVID-19 vaccination form” phishing website underscore the limitations of relying solely on traditional firewall-based defenses. Consequently, there is a growing inclination towards user-centered anti-phishing solutions, exemplified by training games such as What.Hack. But could we proactively notify users in real time when they are on the brink of a scam or when their attention wanes? Our research explores machine learning and eye-tracking to identify email-reading weak spots and gauge a user’s risk of succumbing to phishing lures. We put forth innovative hybrid models, TransMLP Link and TransMLP Hybrid, melding the strengths of both Transformer and MLP. Our method also facilitates consistent interpretation of eye-tracking data across varied email interfaces and displays. Our TransMLP Hybrid model boasts an 88.75% accuracy rate, outperforming the standard Transformer model. Our research points to the future of anti-phishing tools that elegantly combine technological advancements with insights into human behavior.