Detection and Identification of Novel Attacks in Phishing using AI Algorithms

Phishing attacks pose a significant threat to cybersecurity, exploiting human vulnerabilities to compromise sensitive information and undermine trust in digital communication. Currently, anti-phishing techniques that have been predominantly researched and used in software products include list-based and web-content based approaches. While these techniques provide excellent accuracy against previously known phishing attacks, they offer subpar accuracy against zero-day or novel attacks. Acknowledging this gap, our project, “Detection and Identification of Novel Attacks in Phishing using Artificial Intelligence Algorithms,” endeavours to elevate the detection precision for novel/zero-day phishing incidents. Our approach includes the use of visual similarity methods for enhanced detection where it analyses website screenshots. Our research started from a state-of-the-art dataset which we further enhanced by sourcing and analysing reported phishing pages of globally recognized brands from PhishTank, subsequently pairing these with screenshots of the corresponding legitimate websites. This strategic enrichment process aims to capture the sophisticated visual discrepancies between legitimate and phishing web pages, thereby facilitating the training of our model with a more comprehensive and representative dataset. Coupled with dataset enhancement, we have explored a diverse array of algorithms to ascertain the most effective approach for novel phishing detection. Our proposed approach achieved up to 84% accuracy in detection of phishing pages, compared to existing visual similarity based methods.