Improved Detection and Prevention of Phishing URLs Using Ensemble Machine Learning Techniques

Phishing is a cyberattack that uses deceptive emails, websites, or messages to trick individuals into revealing sensitive information like passwords or financial details. It often involves imitating trusted entities to exploit user trust. Researchers have tackled phishing by developing machine learning and deep learning algorithms to analyse patterns in URLs, emails, and website features. Recent advancements include using ensemble methods to enhance detection accuracy and efficiency. However, the literature reveals a gap in effectively combining ensemble learning techniques to overcome individual model limitations. The goal of this research is to enhance the detection of phishing URLs by harnessing the strength of ensemble machine learning. To combat this, the proposed solution involves using a dataset with phishing URLs from a reliable UCI dataset repository, consisting of 134,850 legitimate URLs and 100,945 phishing URLs. It employs multiple machine learning algorithms: Random Forest, Multilayer Perceptron, and Gradient Boosting Machine, each handling different aspects of phishing detection such as URL-based, HTML-based, and derived features respectively. By combining these models through an ensemble approach, the system can mitigate the weaknesses of individual algorithms and improve predictive performance. Key features such as URL length, the presence of HTTPS, the number of subdomains, and 45 other features were extracted for analysis making it total of 47 features after data pre-processing. The results show that the ensemble model outperformed traditional methods, achieving a high level of 99.6% accuracy in detecting phishing URLs.