MODC: A Pareto-Optimal Optimization Approach for Network Traffic Classification Based on the Divide and Conquer Strategy

摘要: Network traffic classification aims to identify categories of or applications network packets flows. It is an area that continues gain attention by researchers due the necessity understanding composition traffics, which changes over time, ensure Quality Service (QoS). Among different methods classification, payload-based one (DPI) most accurate, but presents some drawbacks, such as inability classifying encrypted data, concerns regarding users’ privacy, high computational costs, and ambiguity when multiple signatures might match. For reason, machine learning have been proposed overcome these issues. This work proposes a Multi-Objective Divide Conquer (MODC) model for combining, into hybrid model, supervised unsupervised algorithms, based on divide conquer strategy. Additionally, it flexible since allows administrators choose between set parameters (pareto-optimal solutions), led multi-objective optimization process, prioritizing flow byte accuracies. Our method achieved 94.14% average accuracy analyzed dataset, outperforming six DPI-based tools investigated, including two commercial ones, other learning-based methods.