MalwareInSight: Classifying Malware Images Using Convolutional Neural Networks

Content Overview

This repository contains all notebooks and the final report (presented as a journal paper) for the study "MalwareInSight: Classifying Malware Images Using Convolutional Neural Networks". The research investigates applying convolutional neural networks (CNN) to the malware classification problem by analyzing images, comparing custom architectures with well‑known pretrained models.

Abstract

Traditional signature‑based malware detection methods often cannot keep pace with the rapid evolution of malware. This work builds on Nataraj et al. (2011)’s innovative approach of converting malware code into grayscale images and leverages deep‑learning techniques to improve malware classification capability. Using the MalImg dataset, the study explores the effectiveness of a custom CNN architecture (named MalwareInSight) versus pretrained models such as VGG16 and VGG19. The custom CNN achieves 99.05 % accuracy, surpassing previous benchmarks. The study highlights the limitations of data augmentation and overly complex architectures on malware images. Results support malware image classification as a viable complementary tool to existing methods, with recommendations for expanding datasets, integrating texture‑based pretrained models, and exploring unsupervised learning in future work.

Keywords: malware, cybersecurity, image classification, deep learning, convolutional neural networks