The Research And Implementation Of The Technique In Finding Communications Network Problem Base On Transfer Learning

This paper is based on network anomaly detection of a certain company,to improve network anomaly detection accuracy by improving the traditional classification algorithm based on the characteristics of the abnormal data network.The datasets of network anomaly detection has three characters.Firstly,the data distribution is extremely imbalance and has less abnormal data.Secondly,data is generated everyday,and it changes with the dynamic change of network scale and topological structure.Lastly,because of the great relevance of different network anomaly,the data also has the great relevance.Based on the characteristics of data uneven distribution,this paper presents the UnbalanceAdaboost algorithm.This algorithm is better perform than the EasyEnsemble algorithm and the SMOTEBoost algorithm.Based on the characteristics that different data sets are associated with each other in Network anomaly detection,this paper presents the Unbalance TrAdaboost algorithm.The UnbalanceTr Adaboost algorithm takes account the data distribution imbalance,and the use of related data sets migration study,performs better than UnbalanceAdaboost and TrAdaboost algorithms in dealing with the imbalance data classification and transfer learning.Based on data generated every day,the distribution of data will change.This paper presents a renewal process to classifier,which using the daily feedback data dynamic updating classifier,so that the distribution of the data changes,but classification accuracy of classifiers will not significantly lower.This paper achieve the UnbalanceAdaboost algorithms,the UnbalanceTrAdaboost algorithms and renewal process based on the Weka.Experiments show that the algorithm proposed in this paper perform better than EasyEnsemble,SMOTEBoost and TrAdaboost in dealing with uneven distribution of data sets which have related data sets can migrate study.