Using the ToN-IoT dataset to develop a new intrusion detection system for industrial IoT devices

Using the ToN-IoT dataset to develop a new intrusion detection system for industrial IoT devices

연구 분야: Infrastructure

학회: Multimedia Tools and Applications

With the rapid expansion of the Internet of Things (IoT), an increasing number of IoT devices are becoming interconnected over the Internet, consequently elevating the vulnerability to attacks targeting these devices. Consequently, conducting thorough and high-quality research in intrusion detection within the industrial IoT domain has become imperative. Machine learning is well suited to be used for intrusion detection in IoT scenarios in the face of increasing data volumes. We leveraged the ToN-IoT dataset, utilizing both the IoT device dataset and the Network dataset contained within, to propose a novel approach that integrates multiple datasets for replicating complex IoT scenarios. Pioneeringly, we introduced the use of sine and cosine component cyclic encoding for temporal attributes. Moreover, we address the challenge posed by data imbalance by employing a combination of the Synthetic Minority Over-sampling Technique (SMOTE) for oversampling and an improved version of the redundant-based Tomek link removal technique for under-sampling. We employ various machine learning and deep learning algorithms to construct binary classifiers, categorize data and distinguish between attack data and normal data. The attack data undergoes multiclass classification to identify distinct types of attack, thereby augmenting the attack database. As for the normal data can either be utilized as is or stored encrypted for future use. To comprehensively evaluate the performance of our intrusion detection systems, we used a set of evaluation metrics, including accuracy, precision, recall, and F1 score. We achieved a 100% accuracy in binary classification, while in multi-class classification, we attained accuracy rates of 98.87%, 99.36%, 99.18%, and 99.18% for precision, recall, and F1 score respectively. The evaluation results indicate that our research exhibits superior performance, consistently performing well on several key metrics.