Identifying the type of network traffic has several advantages, such as detecting and preventing applications that violate an organization’s security policy or improving Quality of Service (QoS) and Quality of Experience (QoE) through traffic engineering. To enhance QoS support for Internet Service Providers (ISPs), a fine-grained classification scheme for network traffic is proposed in this paper. Statistical analysis of the throughput patterns of FTP, video conferencing, and video streaming traffic reveal that using new statistical features can be more effective at distinguishing the Internet traffic, especially from a QoS perspective, compared to the features commonly used in the literature, even for encrypted traffic. In this work, machine learning algorithms for classifying the low-latency traffic are trained using combinations of statistical features including the novel trend identification. Experiments are conducted to evaluate the proposed method using large-scale real network traffic data. Results show that our method can classify the particular type of traffic with accuracy of over 97%, and identify the low-latency traffic in the traffic mix with accuracy of 87%.
Paper
Full text
Low-Latency Internet Traffic Identification using Machine Learning with Trend-based Features
Semantic Scholar · Computer Science · 2023
Abstract
Identifying the type of network traffic has several advantages, such as detecting and preventing applications that violate an organization’s security policy or improving Quality of Service (QoS) and Quality of Experience (QoE) through traffic engineering. To enhance QoS support for Internet Service Providers (ISPs), a fine-grained classification scheme for network traffic is proposed in this paper. Statistical analysis of the throughput patterns of FTP, video conferencing, and video streaming traffic reveal that using new statistical features can be more effective at distinguishing the Internet traffic, especially from a QoS perspective, compared to the features commonly used in the literature, even for encrypted traffic. In this work, machine learning algorithms for classifying the low-latency traffic are trained using combinations of statistical features including the novel trend identification. Experiments are conducted to evaluate the proposed method using large-scale real network traffic data. Results show that our method can classify the particular type of traffic with accuracy of over 97%, and identify the low-latency traffic in the traffic mix with accuracy of 87%.