A deep learning IDS solution for IoT infrastructures
Beatrice-Nicoleta Chiriac, Florin Daniel Anton, Radu Nicolae Pietraru, Anca Daniela Ioniţă, Bogdan-Valentin Vasilică · 2025
Networks formed by embedded devices have become so complex that they integrate large numbers of entities which exchange vast quantities of data. If in the past these networks were isolated from the Internet, nowadays this is not applicable anymore and they are much more vulnerable to cybersecurity attacks. Specialized monitoring events systems like intrusion detection systems (IDS) are widely used as a security measure for network supervision. This paper presents such a solution by describing the implementation and the evaluation process of an artificial intelligence (AI) based intrusion detection system for Internet of Things (IoT) infrastructures. The artificial intelligence method that represents the analyzer of the IDS is created by combining the eXtremely Gradient Boosting (XGBoost) algorithm with a deep neural network (DNN) model. The entire model utilizes a comprehensive dataset containing real IoT network traffic captures. The dataset consists of seven different types of common IoT attacks and normal behaviors. During the study, the influence of the quality of the dataset and of the boosting algorithm on the DNN performance was tested. Three different evaluation metrics were used for testing the stacking mechanism, and the model offers a score higher than 0.98 for all of them. The influence of the boosting algorithm on the DNN model was investigated by applying the same data collection to the stacking mechanism as well as to the DNN model without the XGBoost. The research demonstrated that the XGBoost reduced the training time by half and increased the evaluation of metrics values.