Deep Neural Networks for Enhanced Security: Detecting Metamorphic Malware in IoT Devices

Abstract

Today Internet of Things (IoT) has become a key part of the modern world as it enables web-based IoT devices to collect, transfer, and analyze the data of individuals, companies, and industries. IoT provides numerous services and applications via a massive number of interconnected devices and has become an innovative attack vector for cyber-attacks and threats such as malware attacks that are currently regarded as serious dangers to the security of IoT devices and systems. Such threats are sufficient to infiltrate individual private information that inflicts harm to both the financial standing and reputation in an organization. In literature, researchers have used multiple machine learning and deep learning models to tackle this security threat, however, still accurate classification and detection of metamorphic malware in IoT devices remains a challenge. In this article, we used a deep learning model to accurately detect metamorphic malware in IoT devices. We have employed six models including (VGG16, InceptionV3, CNN, ResNet50, MobileNet, and Efficient NetB0 on Malimg publicly available malware image dataset. The Internet of Things (IoT) would benefit from having a method that could identify metamorphic malware. It isn't possible to rely on detection techniques that are fixed or signature-based. Throughout this research, a straightforward technique for carrying out dynamic analysis to comprehend the behavior of code is suggested. To determine if executables are malicious, it is necessary to first measure the behavior of executables and then utilize this information to make that determination. Additionally, the purpose of this study is to create a classifier that makes utilization of deep learning techniques to analyze complicated behavior reports. The obtained results depict that the proposed model achieves a promising accuracy of 99% and F1-score of 97% employed on the standard Malimg dataset as compared to other existing machine and deep learning models.