An Energy Efficient, Robust, Sustainable, and Low Computational Cost Method for Mobile Malware Detection

Abstract

Android malware has been rising alongside the popularity of the Android operating system. Attackers are developing malicious malware that undermines the ability of malware detecting systems and circumvents such systems by obfuscating their disposition. Several machine learning and deep learning techniques have been proposed to retaliate to such problems; nevertheless, they demand high computational power and are not energy efficient. Hence, this article presents an approach to distinguish between benign and malicious malware, which is robust, cost-efficient, and energy-saving by characterizing CNN-based architectures such as the traditional CNN, AlexNet, ResNet, and LeNet-5 and using transfer learning to determine the most efficient framework. The OAT (of-ahead time) files created during the installation of an application on Android are examined and transformed into images to train the datasets. The Hilbert space-filling curve is then used to transfer instructions into pixel locations of the 2-D image. To determine the most ideal model, we have performed several experiments on Android applications containing several benign and malicious samples. We used distinct datasets to test the performance of the models against distinct study questions. We have compared the performance of the aforementioned CNN-based architectures and found that the transfer learning model was the most efficacious and computationally inexpensive one. The proposed framework when used with a transfer learning approach provides better results in comparison to other state-of-the-art techniques.