An Effective Cost-Sensitive Convolutional Neural Network for Network Traffic Classification

The volume, and density of computer network traffic are increasing dramatically with the technology advancements, which has led to the emergence of various new protocols. Analyzing the huge data in large business networks has become important for the owners of those networks. As the majority of the developed applications need to guarantee the network services, while some traditional applications may work well enough without a specific service level. Therefore, the performance requirements of future internet traffic will increase to a higher level. Increasing pressure on the performance of computer networks requires addressing several issues, such as maintaining the scalability of new service architectures, establishing control protocols for routing, and distributing information to identified traffic streams. The main concern is flow detection and traffic detection mechanisms to help establish traffic control policies. A cost-sensitive deep learning approach for encrypted traffic classification has been proposed in this research, to confront the effect of the class imbalance problem on the low-frequency traffic data detection. The developed model can attain a high level of performance, particularly for low-frequency traffic data. It outperformed the other traffic classification methods.