Phishing is among the most worrying issues in a constantly changing world. Because of the rise in Internet usage, phishing has become a new type of data theft This type of cybercrime refers to the theft of private information and violation of privacy by focusing on human vulnerabilities and technical smuggling. URL phishing (Uniform Resource Locators) is one of the most common types. Detecting a malicious URL is a big challenge. This study concentrates on the enhancement of the phishing detection procedure through the utilization of ensemble learning approaches, notably the Gradient Boosting Classifier, CatBoost, and XGBoost algorithms. Leveraging a comprehensive dataset containing examples of both phishing sites and legitimate sites, the study includes comprehensive exploratory data analysis, rigorous data pre-processing, and rigorous model evaluation using cross-validation. The research extends to include importance analysis, using permutation techniques to reveal critical factors that influence the decision-making processes of models. The results demonstrate the effectiveness of ensemble learning in distinguishing between phishing and legitimate entities, The accuracy results reached 98.14% using Gradient Boosting Classifier and cross-validation technique. while providing valuable insights into the key features that lead to accurate predictions. This research advances the subject of cybersecurity by offering a comprehensive comprehension of crowd learning techniques and their useful applications in fortifying defenses against phishing attempts.
