Research On Traffic Analysis Technology Based On Machine Learning

As the scale of computer networks and applications grows exponentially,the importance of network traffic management and analysis is increasing,and network abnormal traffic is also showing rapid growth,and the potential harm is also increasing significantly.The progress of traditional abnormal traffic detection and traffic classification methods is not obvious,and new research methods need to be introduced.Machine learning algorithms are becoming an effective method for complex and growing network traffic problems.At present,most of the methods of machine learning in abnormal traffic detection and traffic classification have problems in that the selection features are too complicated and difficult to compare with each other,which makes it difficult to process the above two problems at high speed,so how to select fewer Features,significantly reduce computational overhead and complexity,and achieve higher accuracy,which is of great significance in the field of abnormal traffic detection and traffic classification research.Thesis focuses on the above problems in network traffic analysis technology,proposes a machine learning-based algorithm,and selects the latest ISCX dataset that truly reflects the actual network conditions for testing,with the aim of selecting fewer features and achieving higher Accuracy.The contributions of thesis mainly include three aspects:(1)thesis proposes an abnormal traffic detection algorithm based on deep neural network,by optimizing the structure of the deep neural network,the accuracy is significantly improved compared to the related research on the same dataset in recent years when fewer features are selected;(2)Thesis is to solve the common problem of the false positive rate based on the deep neural network.Considering different application scenarios,the KNN-based abnormal traffic detection method is proposed to effectively reduce the false positive rate under the premise of ensuring the accuracy;(3)Thesis is to comprehensively compare and analyze the experimental effects of different machine learning methods applied in the field of traffic classification.The theoretical hypothesis of clustering method applied to traffic classification accuracy is verified by experiments.The importance of labeling sample ratio to clustering method is proved by multi-group semi-supervised clustering experiments.The KNN-based traffic classification algorithm is proposed.Compared with the related research on the same dataset in recent years,the accuracy is still maintained when fewer features are selected.