A Generative Neural Network for Enhancing Android Metamorphic Malware Detection based on Behaviour Profiling

Abstract

Malicious software trends show a persistent yearly increase in volume and cost impact. More than 350,000 new malicious or unwanted programs that target various technologies were registered daily over the past year. Metamorphic malware is a specifically dangerous group of malicious software that perturbs its structure between generations. Detecting these types of malware, thus, appear to be more challenging. Recent research demonstrates that Machine Learning (ML) techniques outper-form traditional methods in detecting known and uncategorised malware variants. Hence, this research aims to investigate the use of ML, a Generative Neural Network specifically, for enhancing metamorphic malware detection in Android (the most popular mobile operating system) via augmenting training data. The results show the augmented training data, containing novel samples derived from Deep Convolutional Generative Adversarial Network (DCGAN) and features from metamorphic malware samples, improves the detection performance of unseen meta-morphic malware.