Deep Learning Based Android Anomaly Detection Using a Combination of Vulnerabilities Dataset

• Published in: Applied sciences 2021-08, Vol.11 (16), p.7538
• Main Authors: Namrud, Zakeya, Kpodjedo, Sègla, Talhi, Chamseddine, Bali, Ahmed, Boaye Belle, Alvine
• Format: Article
• Language: English
• Subjects: Access control; Accuracy; Algorithms; android security; Anomalies; Datasets; Deep learning; deep neural network; Learning algorithms; Machine learning; Malware; Mobile operating systems; Neural networks; Security; Smartphones; Source code; support vector machine; Support vector machines; Third party
• ISSN: 2076-3417; 2076-3417
• DOI: 10.3390/app11167538

As the leading mobile phone operating system, Android is an attractive target for malicious applications trying to exploit the system’s security vulnerabilities. Although several approaches have been proposed in the research literature for the detection of Android malwares, many of them suffer from issues such as small training datasets, there are few features (most studies are limited to permissions) that ultimately affect their performance. In order to address these issues, we propose an approach combining advanced machine learning techniques and Android vulnerabilities taken from the AndroVul dataset, which contains a novel combination of features for three different vulnerability levels, including dangerous permissions, code smells, and AndroBugs vulnerabilities. Our approach relies on that dataset to train Deep Learning (DL) and Support Vector Machine (SVM) models for the detection of Android malware. Our results show that both models are capable of detecting malware encoded in Android APK files with about 99% accuracy, which is better than the current state-of-the-art approaches.