Malware Detection Based on Deep Learning

Abstract

In this paper, the author made a dataset and used a deep learning framework to classify malware in the Windows environment. More and more malware is generated nowadays due to the widespread use of the Internet. There has been an enormous increase in malware attacks. To detect the latest malware and keep up with the speed of malware generated. The new method is essential to identify and classify malware samples. Deep learning performed well in classify images. According to deep learning have a good performance in the image. We convert binary files to images. Using deep learning to detect malware. We use the ResNet pre-trained model to train a model for detecting malware files. We converted binary files to gray-scale images and RGB images. Subsequently, used ResNet34, ResNet101, and ResNet152 networks to train a model. The proposed method achieves 98.5294% accuracy in the ResNet101. The author found that using RGB images for training can shorten training time. The accuracy of training using RGB images is only approximately 0.1% worse than using gray-scale images for training. However, using RGB images for training can shorten the time by 9%.