CIRA-Cyber Intelligent Risk Assessment Methodology for Industrial Internet of Things Based on Machine Learning

The Industrial Internet of Things (IIoT) has transformed the industrial control system to Industry 4.0 and plays an indispensable role in improving the efficiency and productivity of industries. The unique features of the IIoT system have potential innovative models in various industrial fields by providing pervasive communication, fast production, efficient data analytics, and well decision-making systems. As IIoT technologies continue to grow, evolve, and mature, cybersecurity threats and adversarial attacks have also increased, compromising the information security triads, i.e., confidentiality, integrity, and availability within IIoT systems. There is a growing need for cybersecurity awareness and a deeper understanding of cyber threats and vulnerabilities in IoT devices to protect valuable data and information. Risk assessment is a proactive approach to identifying, analyzing, and mitigating potential cybersecurity threats. Implementing intelligent and resilient cyber risk assessment is crucial to taking appropriate security measures to avoid cyber threats before exploiting vulnerabilities and turning them into liabilities for IIoT systems. This research proposes CIRA, an intelligent and resilient cyber risk assessment methodology to identify, assess, and mitigate potential cybersecurity risks using leading machine learning models. To select a suitable machine learning model, centralized and decentralized ML models are applied and compared using a synthetic dataset. The results of extensive trials revealed that the decentralized (federated) machine learning models like FXboost and FSVM outperformed in terms of prediction accuracy by 98%, with a low communication cost of 0.29 MB, a low False Alarm Rate, and excellent performance in the ROC curve. Centralized machine models such as XGBoost and LGBM have a prediction accuracy of 99%, less training time, low FAR, and the best ROC value. The performance of the models is evaluated with the state-of-the-art matrices such as precision, recall, and F.1 score has shown that the centralized ML models are feasible for small to medium datasets, however, Federated Learning techniques are outperformed for medium to large datasets to IIoT systems with safety, security, accuracy, and scalability of IIoT nodes.