Optimized Bi-LSTM: a novel approach for attack detection in industrial IoT

The Industrial Internet of Things (IIoT) plays an important role as its implementation requires scalable and secure methods. IIoT integrates databases, machines, various sensors, and services to improve our lives in many ways, including e-healthcare and smarter cities. Accurately identifying attacks in IIoT is critical to mitigating security risks, and the volume of such attacks is rising to an unprecedented level. As IIoT solutions integrate into larger operational systems, additional planning and development of a new mechanism is required to ensure privacy and security. Many deep learning approaches have been developed to focus on intrusion detection. However, such approaches suffer from low detection accuracy, efficiency, privacy, and high time consumption. To mitigate these issues, this study proposes a novel approach called bi-long short-term memory (Bi-LSTM) in IIoT security for privacy and security in industrial systems. To improve the performance of Bi-LSTM, we use modified coot optimization. Our proposed Bi-LSTM approach effectively detects attacks in the IIoT environment. To assess its performance, we analyze various evaluation measures including precision, sensitivity, Matthews correlation coefficient (MCC), and accuracy. The results of our performance analysis reveal that the proposed optimized Bi-LSTM method achieves a higher performance rate of 98.91% accuracy, 95.86% sensitivity, 95.63% MCC, and 97.52% accuracy than other existing approaches. The existing approaches such as CDW_FedAvg, RNN, COSNN, and RCS-DOFH achieved lower accuracy of 98.52%, 97.83%, 95.17%, and 96.24%, respectively. The findings highlight the potential for improved cybersecurity in the IIoT, emphasizing the importance of effective planning and innovative critical systems.