Hybrid Machine Learning Framework for Network Intrusion Detection in IoT-Based Environments

Recent advancements in Internet of things (IoT)-based critical infrastructures (CIs) have contributed towards the generation of massive sensor data. The ubiquity of the sensory devices involved in these infrastructures pose challenges such as safety of devices and their underlying data from attackers, management of secure information flow throughout the network, elimination of unwanted traffic, and preserving quality of service (QoS) of the network. In this view, a hybrid network traffic classification (H-NTC) framework which leverages combination methods to exploit the capabilities of multiple classification models is proposed. The performance of proposed H-NTC model is compared with several baseline predictive models to assess the efficacy of H-NTC model. The proposed H-NTC model is experimentally observed to provide an overall accuracy of 81.7734% with an F1-score of 0.8171, and precision and recall values of 0.8572 and 0.8183, respectively. Further, the receiver operating characteristic (ROC) curves of the proposed model have been comparatively studied alongside the baseline models. The experimental results provided in this study indicate the effectiveness of the proposed H-NTC model and can be successfully validated for classifying network traffic flows for smart CIs.