Evaluating Different Malware Detection Neural Network Architectures

In today’s digital world, cutting-edge technologies in machine learning, cybersecurity, data science, and blockchain are emerging rapidly. Every day billions of devices are connected together through internet for data exchange. Due to the existence of some vulnerabilities present in the device, attackers perform malicious activities through such devices. Detecting the malwares using anti-malware engines and signature-based methods is becoming more and more complex day by day. Various techniques are used by the attackers in the generation of malwares such as encryption, obfuscation, etc., to escape the detection. Few years ago, in traditional machine learning-based malware detection algorithms, feature engineering was the most complex activity to perform. Inefficient extraction of features gives poor performance in the detection of malware samples. Neural network techniques in deep learning give better results in the identification of malware data. This chapter presents a survey on the performance of various deep learning-based neural networks for malware detection and classification. It is also noted that these models require a lot of computational time during training. Transfer learning is a technique used to reduce the training time by using pre-trained models such as VGG-16, ResNet-50, Inception V3, etc. This chapter explains how the evolution of neural networks changed the malware analysis giving exceptional results during detection of malware samples. It classifies, compares, and evaluates several deep learning-based malware detection approaches published in literature elaborating their advantages, disadvantages, applicability, assumptions, and scope.