Improving Phishing Detection Systems Using Kolmogorov-Arnold Networks

Abstract

The phishing activity of cybercrimes accounts to about 30% and is a great threat as over 240,000 cases were documented in 2020 only. Generally, these come in the form of false emails and websites, which aim at stealing essential information that leads to identity theft, financial losses, and compromising through security breaches. Phishing distrusts digital services and largely hits the world economy. This paper uses deep learning architecture capable of processing complex data named Kolmogorov-Arnold Networks (KAN), to set up a phishing detection system. The system achieves more accurate, precise, recall, and F1-score than the typical machine learning models by processing data from URLs, emails, and network traffic and minimizing false positives. The technology is dynamic and it changes all the time, therefore, it is quite good at detecting new phishing maneuvers. Future work towards better real time-detection and enhanced resistance against emerging cyber threats, our work provides a reliable and straightforward method to phishing detection.