PERFORMANCE EVALUATION OF ENSEMBLE LEARNING ALGORITHMS AND CLASSICAL MACHINE LEARNING ALGORITHMS FOR PHISHING DETECTION

ABSTRACT

The advancement in technology and subsequent prevalence of electronic commerce in world trade has resulted in the migration of the majority of banks’ customers to e-banking for their day- to-day financial transactions. This advent of e-commerce has also attracted the attention of fraudsters to defraud bank customers, therefore making e-commerce platforms to be prone to several attacks of which the most common attack is phishing. Therefore, this paper explores the use of ensemble learning and classical machine learning algorithms for the detection of Phishing on e-commerce websites and further evaluates the performance of the models. The three ensemble learning algorithms used are Adaptive Boosting, Majority Voting, and Stacking Ensemble while the three classical machine learning classifiers that were induced include Naïve Bayes (NB), (KNN) K-Nearest Neighbour, and (DT) Decision Tree, to learn phishing patterns in an e-commerce website phishing dataset. The developed models were used to detect phishing in new unseen data instances. The findings of this study show that the accuracy of Naïve Bayes is 89.69%, K- Nearest Neighbour is 93.34%, and Decision Tree is 83.90%. Also, the accuracy of the Stacking Ensemble classifier is 93.94%, that of the voting ensemble is 92.82% and adaptive boosting is 93.63%. These results show that the Stacking Ensemble classifier proves to be the best performing model in detecting phishing in the contest of this study, out of the six models that were considered in the study. The performances of the models are satisfactory and could be adopted in the detection of phishing attacks on e-commerce websites

Keywords: Adaptive Boosting, Decision Tree, K-Nearest Neighbour, Majority Voting Ensemble, Stacking Ensemble, Naïve Bayes, Phishing