“Hybrid Machine Learning Model for Phishing Detection”

Phishing threats have remained a long-standing information security issue for many years, causing billions of pounds in losses both in the United Kingdom and worldwide [1]. The aim of the study was to develop and evaluate the performance of the Machine Learning models that would detect and monitor phishing attacks more accurately. A single dataset with 42 features, a total of 247950 phishing and non-phishing emails was used to develop eight supervised machine learning models. The metrics used in evaluating the models show that the enhanced hybrid algorithm developed from combining two models (decision trees and the random forests) from the trained models generated the best classifier with an accuracy of 96%, precision 98%, f-measure 96%, sensitivity 94%, MCC 92% and ROC 96%. The enhanced hybrid voting model developed was integrated with a Django web application using 13 important features to build an accurate phishing detection and monitoring application. The model is proposed as a novel hybrid model because it demonstrated higher classification capabilities due to its inherent design to deal with complex patterns, overfitting issue and the presence of many features when compared to the other single analysis models in the experiments.