Call Fraud Detection Behavior Analyses Based On Machine Learning

With the fast development of the communication industry and technology,4G network has covered most parts of the country,and the construction of 5G network is also advancing.In the process of large-scale popularization of communication network in the past two years,Telecom fraud is also quietly resurgent.There are many tricks of Telecom fraud,which are impossible to prevent.More and more people have become victims of Telecom fraud.They have suffered both property loss and spiritual hurt,and countless people have break up families or died.At the same time,Telecom fraud has also become a hot topic in the major news media in the society,and the TSP(Telecom Service Provider)are under great pressure.In order to prevent the increasingly spread trend of Telecom fraud,the work direction of ISP is mainly from the management and the technology.The management way is to standardize the card selling behavior of the card selling channels,and prohibit the sale of cards to suspicious users and to cut the phone resource;The technical way can analyze and model the abnormal communication behavior of numbers through data analysis and mining technology,and shut down the predicted highly suspected fraud phone numbers.This paper focuses on technology to detective Telecom fraud,bases on phone owner’s basic information and voice call behavior characteristic data,and uses machine learning in data mining to model and analyze number characteristic data,focusing on solving two problems:First,the imbalance between positive and negative samples.Because the fraud sample data is far less than the normal sample data,this paper compares the three oversampling algorithms SMOTE,Borderline SMOTE and ADASYN through the experimental data,and verifies that the oversampling effect of Borderline SMOTE algorithm is better.The second is the selection of effective training model.By comparing the current mainstream models such as KNN,logistic regression,decision tree,Random Forest and XGBoost,it is verified that XGBoost has a higher recall rate in the prediction of abnormal communication behavior,which can maximize the accuracy of prediction,enable TSP to shut down high-risk phone numbers on time,and to prevent the loss of people’s life and property.