An Attention-Driven Explainable Deep Learning Framework for Multiclass IoT Cyberattack Detection

The rapid growth and development of Internet of Things (IoT) technology have led to an increased risk of cyberattacks such as DDoS, Mirai, Spoofing, and Reconnaissance attacks. The detection of such attacks by traditional intrusion detection systems is not effective in providing accurate results while maintaining explainability. Therefore, this study proposes an attention-driven explainable deep learning framework for effective detection of various types of IoT cyber-attacks. The proposed framework is based on the 1D CNN architecture with the SE attention mechanism to effectively detect various types of IoT cyber-attacks. The proposed framework is tested and evaluated using the CICIoT2023 dataset containing network flows with 47 different statistical features and eight different types of network traffic attacks: Benign, DDoS, DoS, Mirai, Reconnaissance, Spoofing, Web-based, and Brute Force attacks. The proposed framework provides an accuracy level of 99%, proving its effectiveness and robustness in detecting various types of IoT cyber-attacks. The SHAP-based explainability technique provides better explainability and transparency in the proposed framework.