DeepMalware: A Deep Learning based Malware Images Classification
Mehmood Alam, Adeel Akram, Talha Saeed, Sobia Arshad
Abstract
The rapid development in the field of communication and networks has increased the size and complexity of the network. Due to these reasons, many malwares are generated that create a challenges for systems to detect these malwares accurately. Moreover, the presence of malicious software (malware) with the aim of launching various malware files within the network cannot be ignored. Although, there are numerous efforts by the researchers to develop procedures for automatic classification of malware. The methods of manually analyzing malware file is very time-consuming. Lately, deep learning-based methods are being used for the classification of malware. In this paper, we present a rapid and accurate malware classification based on different Convolutional Neural Network (CNN) architectures—including a custom CNN as well as commodity off-the-shelf CNN architectures such as AlexNet, VGG-16, ResNet-50, Inceptionv3 models. This has been demonstrated on benchmark datasets of Malimg dataset, which is consists of malware images that were obtained after conversion of Malware binaries. The trained models allow accurate classification of malware and report a test accuracy of 98.90%.