Multi-Tier Intrusion Detection System for Underwater IoT Wireless Sensor Networks using Hybrid Machine Learning and Reinforcement Learning Approaches

Underwater IoT wireless sensor network (UIoTWSN) is very important for monitoring and gathering real time data in marine environments. This dynamic and complicated nature of these environments leads way to different security threats like listening in, blocking, forging and flooding. Traditional intrusion detection system (IDS) for UIoTWSN is not better at identifying threats due to many false positives which take long time for computation. An adaptive hybrid deep learning framework for efficient attack detection in UIoTWSN to deal with this problem. This proposed method integrated deep learning and reinforcement learning to create flexible and resilient system. A convolutional neural network (CNN) is employed to select the complex patterns for spatial feature extraction in the underwater network data. Then the temporal feature extraction is handled by Bidirectional Long short-term Memory (Bi-LSTM) and it provides the detailed way of how the network behaves over the time. Then the AdamW optimizer is utilized to make sure that the proposed system works efficiently. This hybrid method employs reinforcement learning to change dynamically to new intrusion patterns and changing underwater conditions. Finally, a multi-tier detection method is made to manage the various types of attacks without any problems. Every tier concern on various threats and it employs unique detection methods. This multi-tier detection method identifies the types of intrusion which makes the UIoTWSN more reliable and stronger. Many experiments shows that the suggested framework is better than the existing methods. With 96.7% of accuracy, 94.5% of precision and 95.2% of memory, this method achieves superior performance. These metrics shows that the framework is better at identifying the intrusions accurately when avoiding the false positives and additional task for the system.