JoWUA

Abstract

These days, with the advent of increasing transactions and communications happening over digital platforms, phishing has morphed to assume any online attacks that involve an attacker(s) posing as a legitimate entity, like banks, in most cases. This paper presents a sophisticated defence approach against phishing sites by modelling novel deep learning methods. To make it more transparent, our method employs Random Forest, Extra Trees and XGBoost models, each of which is a machine learning model with an ensemble classifier technique, LIME (Local Interpretable Model agnostic Explanations). The combination of these models, which are understood to handle complex data well, provides high detection accuracy and robustness. Ensemble methods are used to provide a more proper detection solution, which will reduce the false positive rate and false negative rates, so that better trust is maintained with your user base whilst allowing extra reliability of the system. LIME is a significant tool that gives interpretability of the decisions made by models, which in turn can increase users' trust and help developers to continuously improve their systems. Overall, our study underscores the importance of having agile cybersecurity services that can adapt as fast-moving and persistent phishing threats continue to innovate. Using such sophisticated methods provides our system with a robust and future-proof solution, which supports overcoming new phishing tactics, and being able to handle the threats of the digitalised world quickly.