Hybrid Bi-LSTM-Squeeze Model for Attack Detection and Mitigation in Industrial IoT

Abstract

The IIoT has revolutionized industrial operations by enabling seamless connectivity, data-driven decision-making, and enhanced efficiency. However, the pervasive integration of IIoT systems also introduces unprecedented cybersecurity risks, necessitating robust intrusion detection and mitigation mechanisms to safeguard critical infrastructure. Conventional approaches to IIoT security frequently fall short in addressing the dynamic and complex nature of cyber threats, prompting a paradigm shift towards advanced ML techniques, particularly deep learning, for threat detection and mitigation. This paper introduces a novel hybrid detection model aimed at enhancing IIoT security through improved attack detection capabilities. This proposed methodology addresses key concerns such as detection accuracy, efficiency, privacy, and time consumption. Here, propose a comprehensive framework comprising pre-processing, feature extraction, feature selection, attack detection, and attack mitigation. Notably, we employ Improved SMOTE for class imbalance resolution and min-max normalization for data normalization. Feature extraction encompasses info gain, raw, correntropy, and statistical features, while Improved SVM-RFE aids in feature selection. The fundamental component of this strategy is a hybrid model that combines Bi-LSTM and Improved LinkNet for effective attack detection, followed by the utilization of Improved Entropy-based mitigation to eliminate identified threats.