Comparative Study of CatBoost, XGBoost, and LightGBM for Enhanced URL Phishing Detection: A Performance Assessment

Authors:
  • AOAmmar Odeh
    Ammar Odeh
  • QAQasem Abu Al-Haija
    Qasem Abu Al-Haija
  • AAAbdullah Aref
    Abdullah Aref
  • ATAnas Abu Taleb
    Anas Abu Taleb

Abstract
Phishing via URLs involves cyber attackers crafting deceptive websites or emails, mimicking genuine entities like banks or social media outlets. The objective is to dupe users into divulging personal data, such as their passwords or card numbers. This study assesses the potential of machine learning in identifying phishing domains by constructing and contrasting three distinct models. These models, crafted using CatBoost, XGBoost, and LightGBM techniques, are then juxtaposed against prior solutions documented in academic literature. We employed the UCI phishing domains dataset, sourced from URLs, as a performance benchmark for our models. Findings indicate that the model built on CatBoost outperforms its counterparts and also surpasses earlier documented methods.