Toward Improving Internet of Things (IoT) Networks Security Using Machine Learning Based Intrusion Detection System

As the Internet of Things (IoT) expands quickly, network monitoring will confront new security challenges. By 2030, it is expected that fifty billion physical devices will be interconnected to the Internet. However, most IoT devices are vulnerable due to end users and device manufacturers' lack of security awareness. As a result, designing and implementing specialized security measures and mechanisms suitable for the network environment of the IoT is required. Therefore, how to improve IoT network security challenges remains a critical computer security issue. This paper proposes a Machine Learning (ML) based network intrusion detection system (NIDS) to identify IoT network threats. In the first step of this research methodology, four machine-learning algorithms were implemented and compared using the UNSW-NB15 dataset. In the next step, an oversampling approach was performed with ADASYN (Adaptive Synthetic) to tackle imbalanced datasets. Lastly, hyperparameter tuning was proposed for different models to improve performance and enhance efficiency. The experimental results were analyzed and compared in terms of precision, accuracy, F1-score, area under the curve (AUC), and recall. The results were competitive compared with existing works, with an accuracy of 99%.