An Optimized Ensemble Learning Framework for Phishing and Malicious URL Detection

Malicious URL attacks frequently utilize counterfeit websites and legitimate logos to mislead users, presenting a considerable risk in the digital landscape. These attacks can result in the exposure of sensitive information, such as banking credentials and passwords. Although there are existing anti-phishing strategies, cybercriminals continually adapt their methods, underscoring the necessity for a reliable prediction mechanism to safeguard users effectively. Classification serves as one of the techniques for identifying phishing websites. This paper introduces a model designed to detect phishing attacks through the application of various Machine Learning (ML) classifiers. By analysing the “phishing_legitimate_full.csv” dataset sourced from Kaggle, our proposed system incorporates ML models, including Random Forest, CATB, and Naive Bayes, to identify malicious URLs. The dataset comprises 48 features derived from phishing sites and legitimate sites. The methodology encompasses dataset partitioning, model training, prediction, and evaluation to improve accuracy and minimize latency, employing effective feature selection methods to extract relevant features from raw URLs. Ultimately, our system aspires to advance the development of automated solutions capable of effectively classifying and mitigating phishing threats, thereby enhancing user security in the online environment.