Research On Network Traffic Prediction And Business Recognition Algorithm For Data Mining

The booming communication network and the increasing use of Internet bring a large amount of network data.How to extract valuable feature information from it to effectively supervise the network environment has become a research hotspot.The prediction and identification of network traffic has a broad Prospects.With the continuous development of data mining technology,it is widely used in data processing in many fields.Among them,neural network and integrated learning have certain advantages in dealing with complex network data.This thesis uses data mining technology to predict network traffic and identify services.The main research contents include:The trend of network traffic has feature dependence and real-time abruptness,while the traditional network traffic prediction model has weak generalization ability and cannot perform efficient prediction in real time.This thesis proposes a parallel two-dimensional long-and short-term memory neural network prediction algorithm P-LSTM.In this thesis,the long-term characteristics are mined by the hour and date dimensions,the short-term characteristics are extracted by small time scale data,the feature fusion is used to predict the network traffic.In order to improve the convergence rate,the improved particle filter algorithm is used to optimize the parameter training process of the neural network.The iterative method is used to make the prediction model quickly converge to the stable error,and form the final predictive model PF-LSTM.The performance of the model is evaluated from the aspects of prediction accuracy and training rate.The experimental results show that the average relative error of PF-LSTM is reduced by 8% compared with the traditional LSTM algorithm,and has a higher convergence rate.Aiming at the problem that the large number of micro-features of network traffic and complex attributes,traditional business identification methods can not adapt to high-dimensional data.This thesis proposes a network traffic identification method combined with integrated learning.The micro-features of network traffic are reduced by the hybrid feature selection algorithm NDI-RF,which combines the neighborhood discriminant index.In the feature filtering stage,the interaction between features is considered,and the neighborhood discriminant index is used as the judgment index.The clustering idea removes the redundant features,performs feature assignment on the random forest wrapper,and combines the post-sequence search strategy to evaluate the classification effect of each feature subset,and successively selects the features with better classification performance to form the optimal feature subset.The integrated learning strategy XGBoost training classification model is used to identify the network traffic.The experimental results show that the feature selection algorithm NDI-RF can effectively reduce the size of the optimal feature subset,and achieve higher recognition accuracy through integrated learning.