Internet Traffic Classifier Using Artificial Neural Network and 1D-CNN

Abstract

Internet Traffic frequently expands in realm and complexity. Its rate of evolution is dwarfing the ability to manage and control the Network Traffics which will make it very difficult for the ISPs to optimizing network performance and security issues. One of the major issue is Network Traffic Classification, which is the area of major apprehension these days because of the accretion of networks. There are many primeval methods for Internet Traffic Classification like port and payload based classifiers. But many Internet applications do not use standard port numbers, moreover many application use identical port number and payload method uses packet inspection which erodes privacy of user. This lead to the advancement towards Deep Learning(DL) approach for quality Internet Traffic Classification. This paper proposes DNN approach which predict network traffics based on Fully Connected Neural Network(FCNN) and 1DConvolutional Neural Network(1D-CNN) model. The accuracy of proposed solutions on Andrew Moore's dataset is 98.34% for 1D-CNN and 96.97% for FCNN which proves their effectiveness as internet traffic classifier.