Markov Model Based Traffic Classification with Multiple Features

Abstract

Traffic prioritization has recently become more critical and crucial for home Wi-Fi networks due to the increased number of connected devices and applications. While some of these applications are delay sensitive, some have high throughput requirements. Quality of Service (QoS) in Wi-Fi is achieved via differentiation and prioritization of traffic streams, which can be performed successfully as long as the packets can be classified with high precision. As a solution for this problem, this paper presents a new Discrete Time Markov Chain-based traffic classification algorithm, which exploits a multidimensional feature set, named as k-Nearest Markov Component with 3 Dimensions (kNMC-3D). Considering results obtained on two different datasets with current, most popular multimedia applications from different categories, we present the performance of the proposed algorithm, kNMC-3D in comparison to kNMC, two feature extraction based machine learning techniques, Support Vector Machines (SVM) and Random Forest (RF) and a deep learning approach, Auto Encoder with RF (AE+RF). It is shown that kNMC-3D achieves 84.93% and 90.73% accuracy at the application level, 91.13% and 99.17% accuracy at category level on our dataset and a benchmark dataset, respectively. Outperforming the existing methods that focus mainly on feature extraction, kNMC-3D prevents information loss by making use of sequentiality in the traffic, while it improves kNMC by considering multiple features, number of bits, inter-arrival times and number of packets.