Enhancing IoT Network Security with Deep Learning-Based Anomaly Detection

Abstract

The rapid proliferation of Internet of Things (IoT) devices has introduced unprecedented vulnerabilities, with Distributed Denial of Service (DDoS) attacks posing a major threat to the stability and security of IoT networks. This study provides a comprehensive comparison of deep learning models for detecting DDoS attacks in IoT environments. Six models—Gated Recurrent Units (GRU), Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM), Deep Neural Networks (DNN), Support Vector Machines (SVM), and Logistic Regression (LR)—were evaluated on the CICDDoS2019 dataset. Each model's performance was assessed across metrics including accuracy, precision, recall, F1-score, and their ability to detect both normal and attack traffic. DNN demonstrated superior overall performance, achieving the highest accuracy (99.89%) and attack detection rate. GRU emerged as a balanced option for detecting both normal and attack traffic, while CNN excelled in attack-specific detection. The study also highlights lightweight mitigation strategies and analyses the models' throughput, offering insights into their scalability for real-time deployment. These findings provide a foundation for improving DDoS detection systems, ensuring the robustness and security of IoT networks against evolving threats.