Cybersecurity for Internet of Things: Intrusion Detection with Machine Learning and Dimension Reduction

The IoT connects an increasing number of devices to the Internet, simplifying lives but also exposing new vulnerabilities to cyberattacks. The intrusion detection system (IDS) helps detect and prevent attacks on IoT networks. This study aims to develop an intrusion detection and classification system using machine learning and dimension reduction techniques on two datasets. Performance metrics, dataset characteristics, and critical aspects are analyzed. The RF and XGB methods were used to classify attacks in Bot-IoT and Ton-IoT datasets, with and without dimension reduction. Precision, recall, and fl were used to measure classification performance. XGB outperformed in multiclass classification, while RFC excelled in binary classification. The use of PCA reduced computation time for XGB in binary classification but increased it for RFC. XGB showed decreased computation time and a slight performance impact in multiclass classification.