Visual Malware Classification Using a CNN

Malware classification m ethods a re o ften costly, requiring constant retraining and large amounts of computing power in order to support large models that analyze software in numerous ways. To alleviate this problem, this paper builds upon existing research that uses machine learning to analyze grayscale images of the binary code of malware. By using deep learning, the detection of similarities and patterns in the images of malware is a promising approach to malware classification due to its cost effective nature and resistance to many obfuscation techniques. Currently, malware classification is c arried about using static and dynamic malware analysis, being resource-intensive and requiring time to complete. However, through creating a convolutional neural network, or CNN, images of the binary code of malware can be analyzed for patterns identifying them as specific classes of m alware int he M alimg dataset. The implemented CNN architecture was able to achieve $\mathbf{9 7. 6 \%}$ accuracy on the dataset while maintaining its vanilla and basic modifiable structure.