Research On The Netmwork Traffic Identification Technology Based On Semi-supervised Learning

This thesis is supported by the science and technology project of State Grid called Research and Application of Key Technologies of Power Information and Communications Network Traffic Prediction and Pipeline Intelligence,whose research direction is the service-oriented traffic identification and sensing.The main research of the thesis is the study of network traffic identification technologies based on semi-supervised machine learning.According to the fact that network traffic has numerous features,feature selection algorithms are studied and a feature selection algorithm based on information measure named IMSFS is proposed.This thesis puts forward an improved DBSCAN algorithm to solve the problem of choosing rational input parameters of traditional DBSCAN algorithm and the problem of dataset containing a large number of duplicate data.Combining the IMSFS algorithm and the improved DBSCAN algorithm,a semi-supervised DBSCAN traffic identification method named SDBSCAN is proposed.The results of experiment on Moore datasets show that the traffic identification method proposed in this thesis can get highclassification accuracy rate under the situation that the dataset contains a small number of labeled samples.The whole thesis can be divided into five chapters and each chapters' main content is as follows:In the first chapter,the research background and research purpose are introduced and the development trend and key problems of traffic identification are analyzed.At the same time,the architecture of this thesis is describedIn the second chapter,basic knowledge of traffic identification is introduced and the hypothesis and the kinds of semi-supervised learning are analyzed,which lay the foundation for the following research.In the third chapter,feature selection algorithms are studied and a feature selection algorithm based on information measure named IMSFS is proposed because network traffic has numerous features.The effectiveness of IMSFS is verified on Moore dataset.In the fourth chapter,an improved DBSCAN algorithm is proposed to solve the problem of choosing rational input parameters of traditional DBSCAN algorithm and the problem of dataset containing a large number of duplicate data.A semi-supervised DBSCAN traffic identification method named SDBSCAN is proposed by combining the IMSFS algorithm and the improved DBSCAN algorithm.To verify the effectiveness,the test is conducted on Moore dataset.In the fifth chapter,the research work of this thesis is concluded and the direction of further research is pointed out.