Responsible AI‐Driven Optimized Cross‐Hierarchical Structure‐Detail‐Aware Cascaded Capsule Neural Network for Intelligent 6G‐IoT Attack Detection
N Sreekanth, K Padmanaban, R Giri Prasad, L. Guganathan · International Journal of Communication Systems · 2026
ABSTRACT Intrusion detection systems (IDSs) for sixth generation (6G) internet of things (IoT) networks often suffer from low adaptability, high false alarm rates, and limited ability to handle complex and evolving traffic. Traditional deep learning models struggle to capture spatial and temporal attack patterns, and gradient‐based optimizers often get stuck in local minima, reducing detection accuracy. To overcome these challenges, this work proposes an optimized cross‐hierarchical structure‐detail‐aware cascaded capsule neural network (CS2C2N‐GBFIO) for efficient and responsible attack detection in responsible artificial intelligence (AI)–enabled 6G‐IoT networks. Initially, real‐world IoT traffic datasets are collected. The Quality‐aware fuzzy min‐max neural network (QF2MNet) is used to ensure clean, normalized, and multiview feature representation. Subsequently, the Hybrid Divine Religions and Catch Fish Optimization (DRA‐CFOA) algorithm is used to select the features that are the most discriminative and nonredundant. This is followed by the cross‐hierarchical structure‐detail‐aware cascaded capsule neural network (CS2C2N) that learns both structural and fine‐grained features of traffic behavior and Grizzly Bear Fat Increase Optimization (GBFIO) that learns hyperparameters to obtain optimal convergence. The experiments of Real‐Time Internet of Things 2022 (RT‐IoT2022) and Canadian Institute for Cybersecurity Internet of Things 2023 (CICIoT2023) datasets demonstrate superior performance, achieving 99.36% accuracy, 99.43% F1 score, and a false alarm rate (FAR) of 1.30%. The low computational cost, scalability of the model, and adaptability in real‐time make it appropriate to securely implement the model at fog and edge layers of contemporary 6G‐IoT infrastructures.