An effective intrusion detection approach based on ensemble learning for IIoT edge computing

Abstract

The Internet of Things (IoT) has recently gained immense popularity. IoT is a set of devices that compute, communicate with each other, and consist of a vast network. Industrial IoT (IIoT) is the extension of IoT in the industrial fields. It aims to involve embedded devices in the industrial sectors to improve their activities. However, IIoT raises security vulnerabilities that are more harmful than those of IoT. Thus, intrusion detection systems (IDS) are developed to prevent some devastating intrusions. IDS monitors the environment to detect intrusion in real time. This paper designs an intrusion detection approach using ML for IIoT security. The feature selection and dimensionality reduction methods promote the machine learning models' detection rate and accuracy (ACC). To reduce the computational and time costs caused by the dataset's high dimensionality, we propose using Pearson's correlation coefficient (PCC) and isolation forest (IF). The IF is applied to remove outliers, and The PCC is implemented for the feature selection process. Additionally, we used Matthews correlation coefficient (MCC) to study the impact of our proposed model on the imbalanced dataset as the Bot-IoT. The RF classifier is implemented to enhance the IDS performances. For evaluation, we used the Bot-IoT and the wustl_iiot_2021 datasets to evaluate the performance of our model. Our approach has shown remarkable results with 99.99% and 99.12% ACC, 92.17% and 93.96% MCC, and 92.48% and 99.3% AUC scores on the Bot-IoT and wustl_iiot_2021, respectively. Our results demonstrate that the proposed approach has many advantages and superior performances compared with other models.