RGB-based Android Malware Detection and Classification Using Convolutional Neural Network
Asim Darwaish, Farid Naït‐Abdesselam
Abstract
With the proliferation of handheld devices and due to numerous routes for malware creation, the detection of new sophisticated malware becomes a real challenge. In contrary to conventional machine learning approaches, which require feature engineering and source code analysis, we propose here to use a new RGB-based imaging technique for android malware detection and classification. To combat malware threats, our system is built on a static analysis of the android application packaging (APK) file. First, we perform a novel transformation of the APK file into a lightweight RGB image using a predefined dictionary and intelligent mapping. Second, we train a convolutional neural network on the obtained images for the purpose of signature detection and malware family classification. The experimental results on the AndroZoo [1] dataset show that our system can classify both legacy and new malware applications with a high accuracy of 99.37%, a False Negative Rate (FNR) of 0.8%, and a False Positive Rate (FPR) of 0.39%.