Performance Evaluation of Phishing Classification Techniques on Various Data Sources and Schemes

Abstract

Phishing attacks have evolved into a major cybersecurity concern, prompting extensive research to identify the most effective methods for classifying and detecting these deceptive tactics, which aim to deceive individuals and organizations into revealing sensitive information. This project addresses a notable gap in prior research by systematically evaluating various classification techniques under changing data conditions, ensuring that they are not limited to specific datasets or methods, thus offering a broader perspective on their effectiveness in combating phishing attacks. The study conducted assessments on thirteen contemporary classification techniques that are commonly utilized in preliminary research related to phishing. It subjected them to ten diverse performance measures, aiming to provide a comprehensive understanding of their capabilities. The findings of this research contribute valuable insights into the realm of phishing classification techniques, extending the knowledge base beyond what was previously explored in related studies, and ultimately assisting in the development of more effective countermeasures against phishing threats. The project incorporates the Stacking Classifier, a robust ensemble method, combining RF, MLP, and LightGBM models to achieve 100% accuracy in phishing attack classification. A user-friendly Flask-based front end enables easy user testing and performance evaluation. Implemented user authentication ensures secure access, contributing to a comprehensive evaluation of phishing classification techniques across diverse data sources and schemes.