Cyber threat detection in industry 4.0: Leveraging GloVe and self-attention mechanisms in BiLSTM for enhanced intrusion detection

In Industry 4.0, interconnected systems and real-time communication are vital for seamless operations but expose industrial networks to sophisticated cyber threats. Traditional intrusion detection systems often fail to address modern, evolving attacks. This paper presents a novel cyber threat detection approach using a Bidirectional Long Short-Term Memory (BiLSTM) model integrated with GloVe word embeddings and a self-attention mechanism. GloVe captures global co-occurrence relationships in network events, enhancing contextual representation and detection accuracy. To handle class imbalance, random oversampling balances attack category distributions, followed by Principal Component Analysis (PCA) for feature reduction. The model's parameters are fine-tuned using the Single Candidate Optimization Algorithm (SCOA) and Greylag Goose Optimization Algorithm (GLGOA), improving computational efficiency and detection performance. Evaluation on the CIC-IDS-2018 dataset demonstrates superior accuracy, precision, recall, and F1-score compared to state-of-the-art methods. The model effectively detects intrusions and prioritizes high-risk threats, strengthening cybersecurity in Industry 4.0 environments. This adaptable framework can be enhanced to address more complex attack patterns, ensuring robust protection for critical infrastructures.