Pre-Trained Deep Learning Models for Malware Image Based Classification and Detection

In cybersecurity, there are difficulties in detecting malware: the majority of methods use known malware signatures, and it is necessary to spot malware that has never been identified before. Automatic detection and classification of malware is an important approach in the field of cybersecurity. This approach is based on convolutional neural networks (CNN), which have shown powerful performance compared with traditional techniques (static or dynamic analysis, machine learning techniques) in terms of consummation of time and resources. This article presents a critical comparative study of three Pretrained CNN models: Resnet50, VGG16, and XceptionNet, in order to de-termine which offers the best accuracy for the malware image classification using Malimg dataset that contain the malware files represented as grayscale images.