Evaluating the Fidelity and Efficiency of Network Intrusion Detection Systems Via Deep Learning, Machine Learning, and Deep Hybrid Learning in Industrial IoT Devices
The rise of Industry 4.0 has urged scholars to pursue the possibility of integrating these novel technologies with traditional manufacturing practices such as Lean and Six Sigma. However, the main obstacle in the way to achieving such a vision is the concerns arising from the increasing number of cybersecurity threats. Even though there is a sizable amount of research that has already been done on such integration, the cybersecurity concerns regarding that have been mainly overlooked. To this end, this dissertation proposes applying DL (Deep Learning), ML (Machine Learning), and DHL (Deep Hybrid Learning) to classify and detect such threats. First, novel DL algorithms were deployed on three benchmark Network Intrusion Detection Systems (NIDS) datasets to detect cybersecurity attacks. These datasets are known as UNSW-NB15, BoT-IoT, and ToN-IoT, the proposed DL models showed results with high accuracy values. Then a combination of DL and ML algorithms was utilized with a primary objective to propose highly secured and accurate methods for the successful identification of cybersecurity threats through botnet attacks on the N-BaIoT dataset, the results showed high-performance accuracy and proved the models to be a robust approach for threat detection. Finally, and to improve NIDS, new state-of-the-art DHL models were developed by combining ML and DL algorithms; and then deployed to detect anomalies in traffic data of Industrial IoT devices from the TON_IoT dataset. Results showed that the newly proposed DHL models can successfully detect a variety of cybersecurity attacks. In summary, the results substantiate the robustness of the deployed algorithms.