Phishing Detection: Performance Evaluation of Both Ensemble and Classical Machine Learning Models

Abstract:

The majority of bank customers have switched to e-banking for their regular financial activities as a result of the development of technology and the rise of electronic commerce in global trade. The emergence of e-commerce has also attracted scammers looking to swindle bank customers, rendering e-commerce platforms vulnerable to a number of assaults, the most frequent of which is phishing. The application of ensemble learning and classical machine learning techniques for the detection of phishing on e-commerce websites is there-fore explored in this study, and the performance of the models is further evaluated. In order to learn phishing patterns from an e-commerce website phishing dataset, three ensemble learning algorithms—adaptive boosting, majority voting, and stacking ensemble—were used. Also included are three classical machine learning classifiers: Naive Bayes (NB), K-Nearest Neighbour (KNN), and Decision Tree (DT). The created models were utilized to find cases of phishing in new, previously undiscovered data. According to the study's findings, Naive Bayes, K-Nearest Neighbour, and Decision Tree all had accuracy rates of 89.69%, 93.34%, and 83.90%, respectively. Moreover, the vot-ing ensemble classifier's accuracy is 92.82%, the stacking ensemble's is 93.94%, and the adaptive boosting classifier's is 93.63%. From among the six models that were taken into consideration for the study, these findings demonstrate that the Stacking Ensemble classifier proves to be the best-performing model in terms of phishing detection. The models' performances are acceptable and might be used to identify phishing assaults on e-commerce websites.

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