Mobile Network Traffic Identification Based On Multiple Classifier Fusion

Due to the rapid development of the mobile network and the proliferation of mobile applications,mobile traffic identification plays a crucial role in network management and cyber security.Different from the traditional workstation traffic,the particularities of mobile traffic,such as fine-grained identification,encrypted traffic,and the pervasive apps,pose great challenges for traditional traffic identification technologies.At the same time,machine learning has shown promising performance in many classification tasks,and it is quite potential to solve the problems existing in mobile traffic identification technologies.Hence this paper carried out related research on mobile traffic identification technologies based on machine learning.The main work of this paper is as follows:(1)In order to identify the traffic of apps of interest and detect the unknown app traffic,this paper proposes a multi-layer framework based on multiple classifier fusion to identify mobile traffic.First,to avoid the impact of technologies including encryption and tunnel on mobile traffic identification,a traffic feature set which only contains the side-channel data information and few payload bytes of network traffic is designed.Then,a multilayer classifier based on decision trees is used to do fine-grained traffic identification and unknown traffic detection in different layers.Finally,a representative,large-scale mobile traffic dataset is collected to validate the effectiveness of this method.The experimental results show that the proposed method achieves high precision and can detect unknown app traffic effectively.(2)This paper proposes two algorithms to optimize the implementation of decision tree acceleration based on FPGA.Hence the classifiers proposed in(1)can be accelerated on FPGA easier.First,a floating-point number discretization method is proposed to eliminate the floating-point number of decision trees.Then a decision tree pruning method is proposed,which can adaptively adjust the model according to the hardware resource constraints.Thereby avoiding reprogramming FPGA everytime the model is updated.(3)In order to carry out a preliminary coarse-grained identification of pervasive unknown app traffic that cannot be identified in(1),this paper analyzes and summarizes network traffic behavior of apps in service aspect.First,K-Means is used to cluster traffic with similar statistical features.Then Traffic Dispersion Graph is used to visualize the network behavior of each type of traffic.Finally,multiple metrics,including the node degree,connected components number,and other statistical features,are used to characterize the behavior of traffic under different services.The experimental results show that some services traffic can be identified effectively.This paper studies the mobile network traffic identification technologies based on multiple classifier fusion to solve the challenges existing in this field.This paper is of guiding significance and practical value for deploying the mobile traffic identification technologies based on machine learning in the real-world networks.