Hybrid vs Ensemble Classification Models for Phishing Websites

Abstract

Phishing is an internet crime achieved by imitating a legitimate website of a host in order to steal confidential information. Many researchers have developed phishing classification models that are limited in real-time and computational efficiency.  This paper presents an ensemble learning model composed of DTree and NBayes, by STACKING method, with DTree as base learner. The aim is to combine the advantages of simplicity and effectiveness of DTree with the lower complexity time of NBayes. The models were integrated and appraised independently for data training and the probabilities of each class were averaged by their accuracy on the trained data through testing process. The present results of the empirical study on phishing website dataset suggest that the ensemble model significantly outperformed the hybrid model in terms of the measures used. Finally, DTree and STACKING methods showed superior performances compared to the other models.