Implementation And Performance Analysis Of Machine Learning Models For Detecting Phishing Attacks On Websites

Liong, Kah Pong (2023) Implementation And Performance Analysis Of Machine Learning Models For Detecting Phishing Attacks On Websites. [Final Year Project Report] (Unpublished)

	PDF LIONG KAH PONG (24pgs).pdf Download (347kB)
	PDF (Please get the password by email to repository@unimas.my , or call ext: 3914 / 3942 / 3933) LIONG KAH PONG (fulltext).pdf Restricted to Registered users only Download (5MB)

Abstract

In the contemporary world, phishing attacks have become more apparent and caused tremendous financial loss to internet users. When attackers instrument these phishing attacks, an indispensable component frequently used together is a phishing website. Phishing websites are constructed to steal confidential information such as login credentials from victims. Usually, phishing websites are created resembling legitimate sources to deceive victims. To prevent users from falling victim to phishing websites, a machine-learning-based solution is proposed in this project. This project aims to detect phishing websites by implementing a tool that is built from a machine-learning model. This machine learning model is trained using known datasets on phishing websites and legitimate websites. So, features or attributes of phishing websites need to be discovered and this is achieved by looking at techniques phishing websites used to mimic legitimate sources. With the scholarly review of techniques employed by phishing websites, it is decided that they can be identified by their URLs and their SSL certificate information. Then, a machine learning tool is selected to build machine learning models that use three different machine learning algorithms, which are Support Vector Machine, Random Forest, and XGBoost. By having three different machine learning models, performance on how well these models classify phishing websites can be done. With the models successfully trained, they are deployed as a Chrome browser’s Extension and subsequently tested. These models are then evaluated with accuracy, precision, and recall. Finally, the testing and evaluation is done, and XGBoost is proven to be the best performing model in terms of accuracy, precision, and recall.

Actions (For repository members only: login required)

View Item