Improved DDoS Detection Models using Autoencoders and Generative Adversarial Networks for Internet of Things-based Networks

Abstract

Denial-of-service (DDoS) attacks represent the primary threat to the continuous and efficient performance of the Internet. It may have been an element in server delays disconnections, host issues, lost revenue and production, and website vulnerability. Standard machine learning algorithms suffer from increased false-positive rates and reduced rate of detection when new threats develop. Therefore, the DDoS detection devices must include high-performance machine learning classifiers with low false-positive rates and high prediction accuracy. This research paper presents an in-depth study into the scalability and resource efficiency of Deep Learning-based DDoS detection models, specifically convolutional neural networks (CNNs), recurrent neural networks (RNNs), and auto encoders. With a rapid growth of internet of things (IoT) devices, there has been an increase in both the amount and intensity of network attacks. Attacks which result in a denial of service (DoS) or distributed denial of service (DDoS) are considered to be the most prevalent in IoT networks in recent years. Since a majority of current security solutions—firewalls, intrusion detection systems, etc.—filter all valid and malicious data through static, predefined standards, they have been unable to recognize advanced DoS and DDoS attacks. However, when coupled with techniques based on artificial intelligence (AI), these solutions can become reliable as well as effective. We also investigate the effectiveness of transfer learning and model stacking techniques to improve detection performance. Various DDoS scenarios are simulated in a cloud computing environment to assess real-time performance metrics such as latency, throughput, and resource utilization. Experimental results demonstrate that while deep learning models provide high accuracy and F1 scores, the application of transfer learning and model stacking further enhances these metrics. Importantly, we also find that certain architectures demonstrate superior scalability and consume fewer resources, making them better suited for real-world cloud computing applications. The findings from this research contribute valuable insights for the deployment of scalable and resource-efficient DDoS detection systems in cloud computing.