A Novel Approach To Improve The Efficiency of Fake Websites Detection Techniques

Abstract

Phishing is a kind of cyber-attack in which perpetrators use spoofed emails and fraud- ulent web sites to lure unsuspecting online users into giving up personal information. This project looks at the phishing problem holistically by examining various research works and their countermeasures, and how to increase detection. It composes of three approaches. In the first approach, focus was on dataset gathering, preprocessing,features extraction and dataset division in order to make the dataset suitable for the classification process. In the second approach, focus was on metric evaluation of a set of classifiers (C4.5, SVM, KNN and LR) using the accuracy, precision, recall and f-measure metrics. The output of the individual classifier study is used to choose the best performed individual classifier. The final approach is divided into two parts; the first part focus on the increasing detection rate in phishing website algorithm by choosing a suitable design for classifier ensemble and also choosing the best ensemble classifier which will then be in comparison with the best individual classifier. The second part focused on choosing the better of the two part. The resulting outcome of the study shows that the individual classifier method performed better with an accuracy of 99.37% while the chosen ensemble had an accuracy of 99.31%. This result can be attributed to the small size of dataset used as it was shown in past researches that K-NN performs better with a decreasing size of dataset while classifiers like SVM and C4.5 performs better with increasing size of dataset.