The Impact of Data Scaling Approaches on Deep Learning, Random Forest and Nearest Neighbour-Based Network Intrusion Detection Systems for DoS Detection in IoT Networks

In the rapidly expanding realm of the Internet of Things, network security is of paramount importance, especially in the face of an increasing number of DoS attacks leveraging IoT devices. This paper examines the underexplored area of the impact of data scaling approaches on the effectiveness of machine learning-based Network Intrusion Detection Systems in detecting DoS attacks in IoT networks. Specifically, it evaluates the performance of three classifier algorithms, K-Nearest Neighbour, RandomForest, and Deep Neural Networks, on three different datasets, focusing on how distinct feature scaling methods influence detection capabilities. Through a comprehensive experiment, the paper finds that the choice of scaling method can significantly impact the performance of the NIDS. Results vary across datasets and algorithms; for example, the 'Standard' scaling generally outperforms others for ANNs in one dataset, while the 'Quantile' and 'Power' scalings are more effective for ANNs in another. This work fills the gap in the existing research on the machine-learning-based network intrusion detection and has the potential to guide the development of intrusion detection systems, particularly in the complex and vulnerable landscape of IoT.