Demalvertising: Enhanced Detection and Mitigation of Malvertising on Android Devices

Abstract

Web Advertisements are the bedrock for the internet economy, having global ad spending projected to reach $645.80 billion by the close of 2024. However, this blessing can turn into bane as this lucrative industry faces peril from malvertising – the injection of malware or malicious code within web advertisements. This study looked at the application of machine learning techniques to detect malvertising threat, with particular focus on Android devices. The study used the CCCS-CIC-AndMal 2020 dataset which contained a collection of benign and malicious android applications. The study also evaluated several machine learning models such as Decision Trees, Random Forest, XGBoost, Convolutional Neural Networks (CNN). The findings indicated that ensemble method, XGBoost and Convolutional Neural Networks (CNN), showed superior performance in comparison with the other models. Performing machine learning models were integrated with a web-browser extension infrastructure to provide an enterprising detection against malvertising threat. The study discussed limitations and future directions, including the need for a more dynamic feature analysis and potential platform expansion.