Evaluation of Machine Learning Classifiers for Phishing Detection
Date
2016-10-19
Authors
Kazi, Rabail
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
One of the common techniques used by attackers to break security and steal private and confidential information is phishing. An effective way to defend against phishing is to use an add-on filter. However, it is vital for the phishing detection system to be accurate. The phishing detection system used in this project is a website filter based on the Simple Logistic heuristic which is a machine learning algorithm. Weka is a tool used for implementing machine learning algorithms. In this report, several classifiers present inside Weka are tested against a fixed data set. The aim is to examine machine learning classifiers for detection of phishing. Experimental results are presented which demonstrate that Random Forest outperforms all other classifiers with an accuracy of 93%. The accuracy is further improved for Random Forest by using the Auto-WEKA classifier. This classifier is able to detect up to 99% of phishing websites, with a False Positive Rate (FPR) of only 1%. Thus, the accuracy of the phishing detection system can be improved by using the Random Forest classifier and Auto-WEKA.
Description
Keywords
Phishing, Machine Learning, Detection Accuracy