Machine Learning-Based Histogram Boosting Approach for Attack Detection in IoT Networks

The attack detection is crucial in the Internet of Things (IoT) networks for preserving optimal network performance and system integrity. However, existing methods are insufficient to handle the large volume of available data, missing values, and high variance. Moreover, an integrated infrastructure is required to synchronize the various processes involved in attack detection. To address these challenges, we proposed an attack detection framework comprising three phases. The first phase addresses missing values, standardizes data, and maps attack classes based on shared features, reducing the dataset's complexity. In the second phase, oversampling and undersampling methods are sequentially employed to balance the dataset and reduce biases in highly imbalanced datasets. The third phase is a machine learning-based (ML) Histogram Gradient Boosting (HGB) model used to improve the prediction of attacks accurately and efficiently. This three-phase framework enhances ML models' performance and enables them to handle more complex classification tasks. Additionally, we investigate various ML and boosting models to increase the generalization capabilities of the classification. Simulation results conducted on actual IoT datasets indicate that the HGB model outperforms existing ML and boosting models, making it a suitable solution for industrial IoT applications.