Ensemble Classifier on Malicious JavaScript Attacks Detection using Extended Features
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
UNIVERSITI MALAYSIA SARAWAK
Abstract
Description
The rapid expansion of the Internet and rich features of web application has led to many security flaws related to web application. Some flaws are due to poor programming practices while some are intentionally scripted by the attackers behind the scene. Among the reported vulnerabilities by Open Web Application Security Project has ranked cross-site scripting as the second most dangerous vulnerability among top ten vulnerabilities. Cross-site scripting attacks are conducted by using malicious JavaScript's on the client side. Due to the growing adoption of JavaScript's as the scripting language for the World Wide Web has led to the exploitation of cross-site scripting. Detecting these malicious JavaScript attacks with high accuracy on client side has thus becomes imperative. In this research, a detailed and systematic study was performed on the genesis of malicious JavaScript attacks and proposed a unified approach by combining static and dynamic analysis using minimalistic emulation of browser for performing comprehensive characterization of webpage features. This approach leverages features from URL and structural features of webpage to train classification model. A Parteo-optimised majority vote based multi-layer heterogeneous ensemble approach is also proposed and implemented for accurate classification of webpages in malicious and benign c~asses. Evaluation of the proposed approach on a corpus of webpages demonstrates that the approach is precise and effective enough in detecting malicious web pages with an accuracy of 98.74% along with sensitivity and specificity of 97.56% and 98.86%, respectively and also have minimal performance overheads in detection of malicious JavaScript code attacks.