An effective scheme for classifying imbalanced traffic in SD-IoT, leveraging XGBoost and active learning

Abstract

The volume and diversity of Internet traffic are constantly growing due to the simplicity of Internet of Things (IoT) technology, making machine learning-powered solutions increasingly essential for efficient network oversight in the future. The IoT applications prefer stringent but various Quality of Service (QoS). To allocate network resources and offer security based on these QoS, network traffic classification is the foremost solution and a complex part of modern communication. Software Defined Networking (SDN) is combined with machine learning (ML) to automate traffic classification in the IoT network. Nevertheless, uneven class distribution in traffic classification is brought about by the immanent features of Software-Defined IoT (SD-IoT) networks, which could hinder classification performance, particularly for minority classes. In order to solve the issue of class imbalance in SD-IoT environments, this study introduces a Cost-Sensitive XGBoost with Active Learning (AL-CSXGB) algorithm. This unique approach characterizes class distribution from a new point of view. The proposed work dynamically assigns a weight to different applications and actively queries to label new data points iteratively to acquire better accuracy. Experiments on the MOORE_SET and ISCX VPN-nonVPN datasets are used to ensure the efficiency of the algorithm under consideration. The experimental findings show that AL-CSXGB outperforms the other state-of-the-art methods regarding classification accuracy and computation time and alleviates the imbalance problem in SD-IoT networks. The proposed scheme achieves an accuracy of 98.4% on the MOORE_SET dataset and 98.89% on the ISCX VPN-nonVPN dataset, demonstrating its effectiveness and reliability in diverse scenarios.