The Study And Implementation Of Mining Network Log By Bayesian Network

Network is becoming the most popular environment, and it has been developed as ahuge and extensive information database. Under the large-scale database, Vast networkapplications produce huge network logs which might hide abundant information to be mined.To find out these hidden information can improve the network service quality. It is greatlyhelpful to network administration.Bayesian network is one of the most e?ective model in the field of uncertain knowl-edge and reasoning. This thesis applies Bayesian network to network log mining, finding outthe underlying and useful information from network log, in order to make proper decision-making in the administration of networks and make relevant management to improve net-work service quality.This thesis describes how to apply the Bayesian network to mine a real large networklog, which contains several tens million records per day, in order to predict network tra?c.To mine such a huge dataset, this thesis firstly filters and merges the dataset by some criterionto transform the original dataset into the normative dataset for learning Bayesian network.Then the methods based on score metrics with ORSearch algorithm and an incrementalapproach to learning Bayesian network for datasets of di?erent sizes are introduced. Finally,a classification of the network tra?c is conducted by computing the conditional probabilitybetween the variables based on the constructed Bayesian network.Large experimental results show that the accuracy of the prediction of network tra?c isfrom 48% to 80%. The reasons led to such variants might be discretization techniques whenpreprocessing the log sets, learning Bayesian network approaches from large scale dataset,and noise processing of the log sets.However this thesis provides a practical framework for mining network log by Bayesiannetwork.