Analysis And Modeling Of Abnormal Call Recognition Based On Ensemble Learning

The rapid development of communication and network technology has greatly enriched People's Daily life.But cyber security vulnerabilities have created a large number of personal information leakage,which has led to Harassing and fraud calls have spread like viruses in people's lives.To protect people against potential loss caused by abnormal phone activities,many anti-fraud mechanisms have been proposed by researchers.However,these approaches are heavily rely on annotations of original data,while the methods of feature mining and analysis are insufficient.What 's more,most of these methods are passive detection,cannot give accurate prediction in time.Telecom operators have built data centers to store massive records of telecom users' behaviors,these big data provide new opportunities for actively identifying abnormal phones.How to accurately obtain the user behavior fertures representing the user category from the mass data,and construct the abnormal phone recognition model to actively identify the abnormal phone has become a common concern of communication operators and researchers.This paper studies an abnormal phone detection model based on feature mining and ensemble learning.The main work is as follows:1.A method of Data preprocessing and feature extraction analysis of telecom data,DF for short,is proposed to process telecom data and extract,analyze and reduce the dimension of users' historical behaviors.Firstly,the telecom sample data were preprocessed,and then the user information was fully mined from seven dimensions for feature extraction,analysis,and dimension reduction through PCA algorithm to establish a complete set of telecom user behavior features system,named 'TF.2.One method of an abnormal telephone detection based on homogeneous integration is proposed,which is abbreviated DAPHo I.Moreover,a DF+Bagging algorithm model is proposed in DAPHoI method.Firstly,the data were processed by DF algorithm,and then different training sets were formed by Bootstrap sampling.Training base learning model M+b respectively according to the traditional machine algorithm and neural network algorithm,and integrating different Numbers of M+b models by voting method to build the final abnormal phone recognition model.3.One method of an abnormal telephone detection based on heterogeneous integration is proposed,which is abbreviated DAPHeI.The data were processed by DF algorithm in DAPHeI method,and the data were sampled by Bootstrap and SMOTE respectively.Then,the base model M+b and M+s are trained respectively according to the traditional machine algorithm and neural network algorithm.Finally,different integration strategies are adopted to build the model: the first method integrates different quantities and types of M+b base models through the voting method to build the abnormal phone recognition model;The second model is based on the M+s model and adds a meta-learner to build a two-layer framework for abnormal phone recognition model.In this paper,several comparative experiments are carried out on real telecom data.The experimental results demonstrate that the traditional classification model and neural network model can improve the performance significantly when added to our framework.In particular,the accuracy,F1 score and recall rate of FeMELD model in this paper reached 96.6%,96.7% and 98.1% respectively,which can identify abnormal phones actively and accurately.