Robust Intrusion Detection for IoT Networks: an Integrated CNN-LSTM-GRU Approach

Ensuring IoT security is of utmost importance in today's interconnected world. IoT devices are prone to various security threats and vulnerabilities, making it essential to implement robust security measures. Intrusion detection plays an instrumental role in safeguarding network information security. However, traditional machine learning techniques face challenges when dealing with large volumes of data and diverse intrusion classes. Consequently, their detection accuracy becomes inadequate, especially when encountering unknown or novel intrusions. In this paper, we introduce a novel Intrusion Detection System (IDS) using Deep Learning (DL) models. The proposed system integrates three powerful DL models, namely Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU). Different combination sequences are presented to enhance the system's performance in detecting intrusions by leveraging the strengths of each model. We utilized two datasets: Edge-IIoTset that accurately depicts a network traffic environment consisting of IoT and IIoT applications, and NSL KDD that contains simulated non-real traffic data. We employed several metrics, including accuracy, precision, false positive rate, and detection cost, to assess the system's performance. The experimental results show that our model achieves a perfect accuracy of 100% using Edge-IIoTset in binary classification and an accuracy of 99.95% using the NSL-KDD dataset in multi-class classification. Furthermore, it has an improved detection cost compared to single LSTM and GRU models.