Securing Symbiotic IoT in 6G Networks Using a Hybrid MCBA-6GNET Deep Learning Framework for Anomaly Detection

Abstract

The Advent of 6G-Powered Symbiotic IoT (S-IoT) Networks is poised to revolutionize digital ecosystems by enabling distributed intelligence through Edge-Cloud symbiosis for AI-driven automation. However, the integration of large-scale AI models with resource-constrained IoT devices introduces critical security vulnerabilities, as endpoints increasingly serve as vectors for sophisticated cyberattacks, including unauthorized access, data breaches, and systemic disruptions. Traditional security mechanisms, reliant on static rule-based or shallow machine learning models, fail to address the high-dimensional, dynamic nature of IoT-generated data, necessitating advanced solutions for real-time threat detection. This study proposes MCBA-6GNET, a hybrid deep learning framework that synergizes multi-scale spatial–temporal analysis (via EfficientNet, ResNet50, InceptionV3, and BiLSTM) with self-attention mechanisms to secure 6G-enabled IoT ecosystems. The framework employs adaptive data preprocessing, including outlier mitigation, ADASYN-based class balancing, and min-max normalization, followed by hierarchical feature fusion to capture spatial patterns (e.g., packet length variance, TCP flag anomalies) and bidirectional temporal dependencies (e.g., flow inter-arrival dynamics). Evaluated on the ACI-IoT-2023 and RT-IoT-2022 datasets, MCBA-6GNET achieves 99.97% accuracy (99.95% F1-score) and 99.98% accuracy (99.99% F1-score), respectively, outperforming existing methods by up to 17.5% in accuracy while reducing false positives by 99.97%. This research advances secure AI-IoT convergence in 6G networks, offering a scalable blueprint for real-time anomaly detection and laying the foundation for future innovations in edge-based security enforcement, blockchain-augmented trust frameworks, and self-evolving AI models resilient to adversarial cyber threats.