Image based Android Malware Detection System using ResNet50 and VGG16

Attacks on mobile devices, such as smartphones and tablets, have increased as a result of their rising popularity. One of the most significant risks is mobile malware, which may lead to both security breaches and financial losses. Mobile malware will probably continue to develop and spread since it is used to commit different types of cybercrimes on mobile devices. As the Android operating system has gained popularity, mobile malware targets it especially. Users' security is seriously jeopardised by the quick spread of Android malware apps, which also makes it challenging to manually and statically analyse harmful files. Therefore, it is essential to accurately identify and categorise Android harmful files. There has been some discussion of Convolutional Neural Network (CNN) based approaches in this area, however there is still opportunity for performance enhancement. In this work, using two well-known deep learning models, ResNet-50 and VGG16, we present a transfer learning strategy to effectively detect the Android malware files. By converting malicious and benign APK files into grayscale images, the proposed model is trained on the AndroHealthCheck dataset. On the AndroHealthCheck dataset, our model outperforms state-of-the-art works in terms of accuracy, recall, precision, and F1 measures.