Performance Analysis of Selected Machine Learning Algorithms for the Classification of Phishing URLs

摘要

One of the security threats facing onlinecommunities is phishing-based attacks. These phishingattacks come in different ways. It has been argued thatURL-based phishing campaigns are one of the simplestbut very effective techniques that attackers usednowadays to steal sensitive information and robinnocent cyber users of both financial and nonfinancial credentials. However, most of the past studiesfocus on using signature-based approach for detectingmalicious URL instead of Machine Learning basedtechniques. We equally observed that studies that usedML techniques used dataset with limited features ordoes not consider comparative analysis. This workproposed a Machine Learning-based approach forphishing url detection. We investigated theperformances of four supervised machine learningalgorithms in the classification of phishing uniformresource locators. Experimental results showed thatDecision Tree classifier has the overall bestperformance across accuracy, precision, recall and f1-measure used as metrics. This is followed by GaussianNa?ve Bayes’ which also performed very well comparedto the poor results recorded by Logistic Regression andNearest Neighbour algorithm across all the metrics.Thus, Decision Tree classifier was able to classifyphishing url better than all other selected algorithms.