Effective and Intelligent Hacdl Framework for Robust Iot Security Toward Multi-Class Attack Detection Using Capsule Networks

Due to the inability of conventional intrusion detection systems (IDS) to keep up with changing attack patterns, new cybersecurity vulnerabilities have emerged as a result of the fast growth of the Internet of Things (IoT), especially in large-scale networks. Our innovative hybrid Adam capsule deep learning (HACDL) model is designed to tackle this issue and detect intelligent intrusions in IoT networks. Combining Capsule Networks' ability to represent features hierarchically with the Adam optimizer's adaptive optimization efficiency, the suggested model achieves both goals. While traditional CNN-based intrusion detection systems fail to detect complex, multi-class attack behaviors, HACDL succeeds in achieving higher generalization while preserving spatial relationships. The HACDL architecture is tested on the UNSW-NB15 and CICIoT2023 benchmark datasets to evaluate its performance. Impressively surpassing conventional deep-learning methods, our model attains a F1-score of 0.9907 and a 98.7% accuracy on CICIoT2023, as well as a 72.3% F1-score and a 76.5% accuracy on UNSW-NB15. Worms, DDoS, Brute Force, Mirai, and Shellcode are just a few of the attack types for which the model shows strong classification. The efficacy and resilience of HACDL for actual IDS deployment are confirmed by thorough performance metrics, such as confusion matrices, ROC-AUC curves, and per-class precision-recall analysis. Additionally, the model ensures accurate threat detection without overburdening alert systems by displaying low false positive rates (FPR ≤ 0.03). To improve Internet of Things (IoT) security and pave the way for future breakthroughs in cybersecurity using capsule-based deep learning, this study introduces a robust, adaptable, and scalable intrusion detection framework.