LLM Comparison For Phishing Detection Using Machine Learning

Abstract

Phishing is one of the most widely used and dangerous forms of cyberattack. Attackers use fake links (Phishing Links) to steal sensitive information from users who enter the link. As internet usage and digital services increase, phishing techniques are also evolving and becoming more difficult for ordinary users to detect. This study aims to analyze the characteristics of phishing links and find the most effective detection method to identify malicious links. The methods used in this study include collecting datasets of phishing URLs and genuine URLs, analyzing URL features such as URL length, domain name length, title, labels, etc. Then, tests will be conducted using statistical analysis approaches and machine learning algorithms to distinguish phishing links from genuine links. Then we will implement Large Language Model (LLM) to explain how the model got their verdict. The final result of this study shows that a combination of certain URL features can significantly improve the accuracy of phishing link detection. Not only that this study also shows the response quality of different LLM models. It is hoped that this research will contribute to the development of more effective cybersecurity systems and increase user awareness of the dangers of phishing in the digital world.