Enhancing security in 6G-enabled wireless sensor networks for smart cities: a multi-deep learning intrusion detection approach

Introduction Wireless Sensor Networks (WSNs) play a critical role in the development of sustainable and intelligent smart city infrastructures, enabling data-driven services such as smart mobility, environmental monitoring, and public safety. As these networks evolve under 6G connectivity frameworks, their increasing reliance on heterogeneous communication protocols and decentralized architectures exposes them to sophisticated cyber threats. To secure 6G-enabled WSNs, robust and efficient anomaly detection mechanisms are essential, especially for resource-constrained environments. Methods This paper proposes and evaluates a multi-deep learning intrusion detection framework optimized to secure WSNs in 6G-driven smart cities. The model integrates a Transformer-based encoder, Convolutional Neural Networks (CNNs), and Variational Autoencoder-Long Short-Term Memory (VAE-LSTM) networks to enhance anomaly detection capabilities. This hybrid approach captures spatial, temporal, and contextual patterns in network traffic, improving detection accuracy against botnets, denial-of-service (DoS) attacks, and reconnaissance threats. Results and discussion To validate the proposed framework, we employ the Kitsune and 5G-NIDD datasets, which provide intrusion detection scenarios relevant to IoT-based and non-IP traffic environments. Our model achieves an accuracy of 99.83% on the Kitsune and 99.27% on the 5G-NIDD dataset, demonstrating its effectiveness in identifying malicious activities in low-latency WSN infrastructures. By integrating advanced AI-driven security measures, this work contributes to the development of resilient and sustainable smart city ecosystems under future 6G paradigms.