Abstract
The adverse effects of phishing attacks on unsuspecting victims are damaging and nefarious. Stealing of information from unsuspecting users surges on the internet and various solutions have been proposed to curb this menace. Apparently, the evasiveness of phishing attacks through dynamic processes renders these solutions ineffective. To curb this prevalence, machine learning (ML)-based solutions are developed and deployed as it offers continuous learning of phishing dynamics as opposed to explicit or static countermeasures. However, existing ML solutions suffer drawbacks in the case of high false alarm rates and relatively low accuracy values. Hence, this paper proposed novel intelligent tree-based ensemble approaches for phishing website detection. Particularly, ensemble methods (ABELM, BAELM, MABELM) are developed based on Naive Bayes Tree (NBTree) and Best-First Tree (BFTree) classifiers. NBTree uses an NB classifier at the leaf nodes of a decision tree while BFTree deploys the best first induction method to add the best split in each step of the decision tree. Experimental results showed that the proposed methods are highly effective for phishing website detection outperforming baseline classifiers and ML-based phishing models from recent studies. Consequently, the tree-based ensemble approaches are viable methods that can be used for detecting phishing websites with dynamic traits.