A Deep Learning-Based Hybrid Framework for Sinkhole Attack Detection in Underwater Wireless Sensor Networks

Underwater Wireless Sensor Networks (UWSNs) play a pivotal role in oceanographic data collection, surveillance, and marine research. However, their unique operating environment and resource limitations make them vulnerable to security threats, notably Sinkhole attacks. This paper introduces a deep learning-based hybrid framework to detect and mitigate such attacks by integrating Graph Neural Networks (GNN), Autoencoders, and Gaussian Mixture Models (GMM). The proposed system leverages graph-based spatial learning, feature compression, and probabilistic anomaly detection to identify malicious patterns with high precision. Designed specifically for underwater conditions, the approach ensures robust intrusion detection while maintaining network efficiency and reliability. This work aims to strengthen trust and resilience in UWSN communication against deceptive routing behaviors.