Research On Abnormal Behavior Detection Based On User Network Data Fingerprint

As internet coverage is prevalent globally,without being noticed,the network is transforming the everyday life of each and every individual.There are a variety of data circumstances and characteristics in the immense data on users' network behavior.By analyzing the user's network behavior data,researchers may observe users' Internet habits and preferences,and then discern the abnormal users,whose deeds may tramatize their own physical and mental health,in a worse case scenario may even destroy the healthy ecological circle of the network environment,together with the fairness and justice of the network transaction.For the sake of the harmonious and healthy development of network ecology,it is essential to identify the users who have abnormal Internet behavior.However,the bulk of massive data poses a challenge to contriving approaches to anomaly detecting.Therefore this paper will analyze the two specific anomalies under the current Internet environment,and complete the scheme design of detecting these two kinds of anomalies.The paper is innovative and contributive in the following dimensions:(1)Integrating DPI(Deep Packet Inspection)Data so as to analyze abnormal behavior patterns from the perspective of user network behavior.Taking into account the whole process of Internet utilization,the paper constructs the network data fingerprint system to demonstrate users'network behavior through the processing of operators' large-scale DPI Data and extraction of effective content.(2)Analyzing the features of various types of abnormal users'behavior.Designing a more sophisticated and accurate anomaly detection scheme on the basis of feature engineering and machine learning model.Objects of the scheme are two kinds of abnormal behavior patterns,scalping users and poor internet habits users.(3)The RPC A(Robust Principal Component Analysis)algorithm is introduced into the feature engineering system to demonstrate the validity of applying the algorithm to the anomaly detection task,meanwhile enriching the special collection and enhancing the capability of feature selection.(4)In the learning model,a refined iForest(Isolation Forest)algorithm,the w-iForest algorithm is proposed.The difference in detection effectiveness of each component of iForest is consequently charted.iForest algorithm is applied to poor internet behavior users so as to prove that w-iForest algorithm performs favorably.