Research On False User Recognition Of Platform Based On Model Fusion

The Internet has grown rapidly during these recent years and many other industries derived from it,In Internet plus environment,many industries combine with the internet to achieve industry innovation and development.Therefore,a lot of companies use some optimization policies to attract users and expand the pr omotion company platform in order to fight for more resources of users.As a result,some new registered users can enjoy the preferential policies of the platform,such as receiving 4G free data getting 20 Yuan cash back and so on.This has bred some gray areas of the so-called Tares Wool party who drills loopholes,batch registration of users of the platform for accessing to benefits.However,the batch of numbers is actually called the false user,the user number is not a normal number,their communication behaviors are different from the normal user's.In order to avoid those losses,the platform joint communications operators to work together to identify tares wool customers to reduce the loss of the platform.Machine learning.uses mathematical statistics knowledge and algorithm theory to establish an effective learning model,mining the internal rules from the data.For the information dimension is not wide,it will not involve user privacy and it will cope with the dynamic changes in the data environment,so machine learning method is very suitable for the recognition of platform false user.This thesis mainly studies the false user recognition based on machine learning.It uses machine learning to identify the user type,uses the user's communication b ehavior and the behavior statistic characteristic of the platform to establish the classification model.The focus of the study is as follows :Data quality will have a profound influence on the effect of classification model while selecting the efficient f eature set from the large number of users ' communication behavior data and Platform.What's more,it is crucial to establish the well classification model.The main contents of this thesis are as follows:Firstly,introduce the application background of false user identification and explain that transforming practical problems into mathematical problems for data mining.Secondly,from the dozens of characteristics of user data,choose good features and delete some redundant features.Thirdly,introduces some simple basic machine learning method and carries out experiments to analyze the effect of the model.Fourth,in the research of ensemble learning,elaborate the learning principle of Ada Boost,random forest,lifting tree and carries out experiments to analyze the effect of the model.Fifth,stacking fusion is used to integrate the integrated learning algorithm and some simple classification algorithms,such as LR and KNN,so as to achieve better of accuracy of user identification.The research can effectively choose the user's behavior characteristics,combine with the model fusion algorithm,improve the recognition accuracy and recall rate,identify false users,thus reducing the business platform and communication operators loss.