Research And Application Of Mobile Communication Traffic Service Prediction Model

The data stored in the communications industry is mainly divided into economic analysis data and signaling side data,with high dimension,diversity and mass.How can telecom operators extract customer groups with traffic requirements from big data,become an important component of traffic management and enhance customers' flow viscosity.Constantly updating the traffic business products,the asymmetries of the traffic products information will make users make choices that do not meet their actual traffic needs.How to select the reasonable traffic business for different segments of customers becomes an important subject in the traffic management practice.This paper adopts the Cross-Industry Standard Process for Data Mining.Firstly,Analyzing business problems to determine the actual demand of traffic business.In order to effectively carrying out mobile communications traffic business prediction research,basing on the actual total consumption and business knowledge of the mobile customers,the mobile customers are divided into three different value segments.Using the analysis method based on RFM model,a scientific and reasonable,quantifiable and standardized customer labelling system is established.Using data mining technology of association rule to study the traffic business portfolio of customers' preferences in the communication industry,the analysis conforms to that the traffic business portfolio that meeting their actual traffic needs.According to the historical data of customers in the communication industry,the traffic business portfolio that meet needs is extracted,and a reasonable assessment is carried out.A logistic regression algorithm based on step by step feature selection method,decision tree algorithm and back-propagation neural network algorithm is implemented,in order to output the possibility of the traffic business combination that customer will handle,and improve the effect of model application.Based on the weight of evidence,information value and the correlation coefficient of variables,the method for selecting classification models of different predictors is carried out.The method can be used to select the feature combinations of the model when ensuring the high impact of the model and to shorten the fitting time of the model,that the modelers can devote more energy to the customer sub-groups,derivative variables and feature selection.The results of the model are validated and the performance analysis is carried out to confirm the stability of model classification prediction results.The classification prediction results of the model can meet the traffic demand of the customers and the expected demand for accurate marketing.