A Deep Learning Approach to Image-Based Malware Analysis

Abstract

Malicious software also referred to as “Malware” is one of the serious threats on the Internet today since it has been growing exponentially over the last decade according to research, causing substantial financial trouble to various organizations. Different security companies have been proposing different techniques to defend from this threat which is a major challenge on the complexity and growing volumes. Recently, malware communities and researchers have begun to apply machine learning and deep learning model to detect potential threats. We propose a malware classification model that takes advantage of the potential of deep learning (DL) models using the convolutional neural network (CNN) and combination of machine learning classifier with CNN such as support vector machine (SVM) for classifying their families. Detection of newly released malware using such models would be possible through mathematical function. That is, $f{:}n \to z$, where n is the given malware and z is their corresponding malware family. Malimg dataset is used to perform the experiment which contains malware image of 25 malware families and 9339 malware samples. CNN has outperformed the CNN-SVM with a test accuracy of 97.5%.