Deep Convolutional Neural Networks Network with Transfer Learning for Image-Based Malware Analysis

Abstract

The complexity of classifying malware is high since it may take many forms and is constantly changing. With the help of transfer learning and easy access to massive data, neural networks may be able to easily manage this problem. This exploratory work aspires to swiftly and precisely classify malware shown as grayscale images into their various families. The VGG-16 model, which had already been trained, was used together with a learning algorithm, and the resulting accuracy was 88.40%. Additionally, the Inception-V3 algorithm for classifying malicious images into family members did significantly improve their unique approach when compared with the ResNet-50. The proposed model developed using a convolution neural network outperformed the others and correctly identified malware classification 94.7% of the time. Obtaining an F1-score of 0.93, our model outperformed the industry-standard VGG-16, ResNet-50, and Inception-V3. When VGG-16 was tuned incorrectly, however, it lost many of its parameters and performed poorly. Overall, the malware classification problem is eased by the approach of converting it to images and then classifying the generated images.