Secure and optimized intrusion detection scheme using LSTM-MAC principles for underwater wireless sensor networks

Underwater Wireless Sensor Networks (UWSNs) are the type of WSNs that transmit the data through water medium and monitor the oceanic conditions, water contents, under-sea habitations, underwater beings and military objects. Unlike air medium, water channel creates stronger communication barriers. In addition, the malicious data injection and other network attacks create security problems during data communication. Protecting the vulnerable UWSN channel is not an easy task under critical water conditions. Many research works proposed in the literature used cryptography principles and intelligent intrusion detection systems to secure the network activities from malicious nodes. However, the need for Machine Learning (ML) and Deep Learning (DL) associated Medium Access Control (MAC) principles is expected for handling the barriers in uncertain UWSN. In this regard, this article proposes a new Intrusion detection system with Integrated Secure MAC principles and Long Short-Term Memory (LSTM) architectures for organizing real-time neighbor monitoring tasks. The proposed system implements Generative Adversarial Network (GAN) driven UWSN channel assessment models and Secure LSTM-MAC principles to protect the data communication. In this regard, the proposed model creates the Intrusion Detection System (IDS) using trained distributed agents. These agents run in each legitimate sensor node contain novel LSTM-MAC engine, intrusion dataset, rule-based monitoring techniques, Secure Hashing Algorithm-3 (SHA-3), Two Fish algorithm and packet filtering tools. The proposed LSTM and agent-based model drives adaptive MAC channel operations to avoid malicious traffics in to legitimate nodes. In addition, this work implements neighbor-based packet monitoring, signal jamming and alert messaging procedures to build reliable security services against different types of attacks. The experiments and the observations reveal the performance of proposed techniques is proved to be 5% to 10% higher than existing techniques in various aspects measured with different metrics.