Enhancing Machine Learning-Based Webpages Phishing Detection with Heterogeneous Distributed Ensemble Feature Selection

The analysis of webpage phishing is a standout computational method used to distinguish phishing from authentic websites through URL examination. In data mining, classification is crucial for constructing machine learning models that facilitate accurate decision-making. Phishing, a significant cyber menace, has recently escalated, inflicting substantial financial damage on individuals, corporations, and the global economy. Phishers typically replicate legitimate websites to trick unsuspecting users and extract vital information. To enhance cybersecurity systems, it's vital to selectively eliminate irrelevant or redundant features, retaining only pertinent attributes to narrow down the search scope. Not all data features are critical for classification, and feature selection can be a complex task. This research introduces a new ensemble feature selection method called Heterogeneous Distributed Ensemble Feature Selection (HDEFS). The authors evaluated and contrasted the features derived from each component of the ensemble framework against baseline features in chosen models. Conducted experiments revealed that HDEFS enhanced the performance of both traditional and ensemble models, also simplifying model complexity relative to earlier studies. The boosted_SVM surpassed all other boosted models, achieving a 97.4% accuracy rate followed by boosted_LR with a detection rate of 94%.