Hybrid Data-Driven Learning-Based Internet of Things Network Intrusion Detection Model

Early and effective intrusion detection plays a significant role in protecting the privacy within and between the different physical devices and entities that are interconnected in Internet of Things (IoT) networks. However, as intrusion techniques and methodologies become rather sophisticated in recent times, traditional detection methods have shown limitations in effectively identifying modern-day vulnerabilities, intrusion and attacks. Data analytical models including deep learning algorithms have been considered and proposed as solutions for IoT networks’ intrusion detection as they have the capacity to recognize patterns in IoT network datasets. However, the voluminous data and class imbalance problem owing to users’ dynamism of IoT network applications have contributed significantly to poor performances achieved by the various proposed models. In this paper, a robust hybrid model consisting of gated recurrent unit and long short-term memory as base learners, with a multilayer perceptron neural network as the meta-learner is proposed for detecting intrusion in IoT networks. To enhance the performance of the proposed model, a feature selection technique is deployed during preprocessing. The proposed model’s performance was evaluated using two modern-day benchmark IoT datasets. According to the results achieved, the proposed model performed better across all metrics considered in comparison with other similar models.