Abstract

The widespread use of the Internet of Things (IoT) has led to significant breakthroughs in various fields but has also exposed critical vulnerabilities to evolving cybersecurity threats. Current Intrusion Detection Systems (IDSs) often fail to provide real-time detection, scalability, and interpretability, particularly in high-speed optical network environments. This research introduces XIoT, which is a novel explainable IoT attack detection model designed to address these challenges. Leveraging advanced deep learning methods, specifically Convolutional Neural Networks (CNNs), XIoT analyzes spectrogram images transformed from IoT network traffic data to detect subtle and complex attack patterns. Unlike traditional approaches, XIoT emphasizes interpretability by integrating explainable AI mechanisms, enabling cybersecurity analysts to understand and trust its predictions. By offering actionable insights into the factors driving its decision making, XIoT supports informed responses to cyber threats. Furthermore, the model’s architecture leverages the high-speed, low-latency characteristics of optical networks, ensuring the efficient processing of large-scale IoT data streams and supporting real-time detection in diverse IoT ecosystems. Comprehensive experiments on benchmark datasets, including KDD CUP99, UNSW NB15, and Bot-IoT, demonstrate XIoT’s exceptional accuracy rates of 99.34%, 99.61%, and 99.21%, respectively, significantly surpassing existing methods in both accuracy and interpretability. These results highlight XIoT’s capability to enhance IoT security by addressing real-world challenges, ensuring robust, scalable, and interpretable protection for IoT networks against sophisticated cyber threats.