An SDN-Orchestrated Artificial Intelligence-Empowered Framework to Combat Intrusions in the Next Generation Cyber-Physical Systems

Automated communication within heterogeneous connectivity of participating devices is an exceptional idea that has allowed the world to take modernized network communication for granted. Internet-of-Things (IoT) is a cutting-edge technology that has enduringly adopted this phenomenal trend and proven its efficiency by facilitating every aspect of human life. The application areas of IoT are widespread encompassing every momentous sector of our life such as the industrial sector, and are touted as a sterling choice for revolutionary communications. However, the same convenient configuration and easily acquirable nature of IoT also make it vulnerable to a divergent sphere of cyber-attacks that demand significant security countermeasures. However, the extensive connectivity of heterogeneous devices in resource-constrained industrial networks warrants the way to design a compatible, robust, and economical security framework against such crucial anomalies. The remarkable acclaim of deep learning (DL) in collaborative integration with software-defined networks (SDN) motivates us to design an efficient anomaly detection mechanism. In this research study, we have proposed a DL-driven SDN-inspired adversaries detection framework (Cu-BLSTMGRU) for IoT-based resource-constrained industrial networks. The proposed model is trained on the CICIDDOS2019 dataset, with its performance being validated on legislative performance metrics in terms of accuracy, precision, recall, and F-score. We have further drawn a performance comparison between Cu-BLSTMGRU and two influential DL algorithms, namely Cu-GRU and Cu-LSTM, along with the recent detection mechanism from the literature. For a comprehensive performance analysis, the designed adversaries detection framework is further compared to some state-of-the-art intrusion detection schemes from the literature. The proposed Cu-BLSTMGRU scheme has outshined the benchmark models and recent detection schemes for all standard evaluation metrics with an attack detection accuracy of 99.65%, precision of 99.19%, recall of 99.12%, and F-score of 99.16%. © (2024), (Korea Information Processing Society). All Rights Reserved.