Malware Classification using Deep Learning Techniques

Over 2.8 billion malware attacks struck in first six months of 2022, affecting everything from small businesses to large-scale corporations. The threat landscape has evolved from mischief to severe cybercrimes and espionage. Therefore, a defence for malware detection and classification is required. Portable Executable (PE) files or malware binaries were collected from dataset comprising of 9339 files of 25 different malware families, which were visualized into gray-scale images. On visualizing, we observed that texture and layout of images of same malware families emerged similar. In this paper, we compare the accuracies of our CNN-LSTM model with 3 pre-trained CNN (Convolutional Neural Network) models- ResNet50, VGG19 and Xception and a CNN model, by classifying the malware images into 25 different families. We transform the binary malware files to grayscale images and run them through a deep learning framework for malware detection and classification. The ability of CNNs to learn the features of these images may lead to the timely and accurate detection of malware. Results show that the CNN-LSTM model predicts classes with a training accuracy of 98.04 %.