Traffic Classification based on Incremental Learning Algorithms for the Software-Defined Networks

A new era of network administration has been ushered in by recent developments in software-defined networks (SDN) and traffic classification (TC) using machine learning (ML) techniques. All network devices are centrally managed and accessible through the SDN, which can simplify the TC process. The traditional data mining/ML approach that uses TC assumes that all of the task's data is always accessible and can be viewed simultaneously without processing time and memory restrictions. Therefore, these approaches are not effective in the case of stream learning since, in more realistic settings, the data is not available all at once and has a distinct distribution. Consequently, incremental learning algorithms (ILAs) can handle online data mining. This study's primary goal is to contrast various ILA approaches to enhance SDN's TC performance. In this study, we propose four ILAs: the self- adjusting memory coupled with the k Nearest Neighbor (kNN) classifier (SAMKNNC), the very fast decision rules classifier (VFDRC), the extremely fast decision tree classifier (EFDTC), and the streaming random patches ensemble classifier (SRPC). Both real and synthetic datasets are used for validation. Experimental findings reveal that the proposed techniques perform better in SDN traffic classification since they can effectively identify drift and use less memory and time.