A Deep Learning Model for Malware Multi-Class Classification based on Colored Malware Images

Abstract

A malicious computer program and its unique attacks have been a source of concern for decades and a major threat to the people of the cyber world. There has been a dramatic increase in malware attacks, their exploration, and the complexity of code and types, which has made malware classification very difficult. With the advent of automated strategies and tools for producing malware, a newly developed malicious program evades detection strategies. Deep Learning (DL) has gained a lot of attention, popularity, and performance in malware analysis. Although DL models reach high-performance levels, they require extensive training samples, high-resolution images, and deep DL structures. This study investigates and highlights the performance of the Convolutional Neural Network (CNN) based malware classifier on colored malware images. To test the CNN model, a well-known malimg dataset was used with a classification speed of fewer than three milliseconds per step, achieving 98.394% test accuracy. The test results encourage the usage of color image processing compared to grayscale images and demonstrate good efficiency and accuracy. It emphasizes that even with basic DL structures, remarkable performance can be attained when dealing with low-dimensional images.