Hybrid Deep Learning-Based Ensemble Model for Detecting Network-Based Cyber-Attacks of Industrial Internet of Services

The Industrial Internet of Things (1IoT) and Industrial Internet of Services (1IoS) have been positioned as crucial pillars of Industry 4.0 and are continuously accelerating in the digital era. Moreover, in the recent era, the focus has shifted from simply connecting systems and smart devices to offering a more evolved and service-oriented architecture that allows the effective utilization of industrial resources. However, the dynamic complexity introduced by IloS and IloT opens the door for cyber-attacks and reveals the industrial architecture of malware and cyber threats. Recent studies have shown that hybrid deep-learning models can potentially mitigate sophisticated cyber-attacks. However, existing models suffer from some concerns, such as low detection rates, false alarms, and resilience to predefined signatures and rules, making them ineffective in detecting sophisticated cyber-attacks. Considering this, this study proposes a deep-learning-based ensemble model (DLEM) that is optimized using ensemble-based models. In DLEM, the hybrid model, which is optimized using ensemble algorithms, is useful for detecting network-based attacks because it utilizes spatial and temporal features to analyze complex and sophisticated patterns. In this DLEM, multilayer perceptron (MLP) captures complex features, whereas Bidirectional Long Short-Term Memory (BiLSTM) uncovers correlations and trends that make this model practical for detecting network attacks that evolve gradually. Moreover, the ensemble model in this DLEM improves the final accuracy by integrating decisions from multiple models, which makes the model more robust against a wide range of attacks and prevents false-negative results. The proposed DLEM with 97 % accuracy is valuable for both practitioners and researchers because it offers a robust framework for industrial defense.