Network Performance Anomaly Detection And Prediction Based On Machine Learning

With the rapid development of Internet technology and the continuous expansion of the network scale,the Internet has become an important part of everyday life.Network multimedia services,network applications,and network users demand higher quality of network communication services.However,the probability of problems such as poor performance and network security risks in the communication networks has also greatly increased.For example,network congestion,network failure,natural disasters and other unexpected conditions affect network performance.Therefore,research on network performance monitoring,prediction and anomaly detection has gained importance.By monitoring large-scale network data streams,network conditions and network failures can be understood in time to optimize network structure and improve network performance.This thesis focuses on network performance monitoring platform,and obtains end-to-end network performance data through Ping ER network performance monitoring platform.It studies the abnormality detection and prediction of network performance.The main contributions of this thesis are as follows:1.This thesis introduces the research background of network performance anomaly detection and prediction and the Ping ER framework,including the monitoring mechanism and indicators of Ping ER,and points out the problems of abnormal values in Ping ER.The network performance monitoring in IPv6 environment is realized by ICMPv6 protocol,and the experimental data is collected successfully.2.Research the network performance anomaly detection based on neural network.This thesis introduces the depth feedforward neural network,basic unit,error inverse propagation,optimization function and so on in detail.An anomaly detection model based on feed-forward fully connected neural network is proposed.The validity of the model is verified by calculating its accuracy,false alarm rate and comparing with clustering based outlier factor algorithm.3.The research on network performance data to predict,the use of multiple linear regression algorithm,random forest algorithm,Gradient Boosting,XGBoost network forecast model is set up respectively,and then using the root mean square error(RMSE)algorithm compares the different model for network performance prediction accuracy.The experimental results show that the model of multiple linear regression improves the accuracy of network performance prediction,realizes the accurate prediction of network traffic,and has strong practical value.This thesis studies the characteristics of network performance data,uses appropriate machine learning algorithm,establishes data analysis model,and realizes abnormal detection and prediction of network performance to provide effective data support for network users and network managers in terms of improving network performance and optimizing network structure.