The Research Of The Telecom User Churn Prediction Model Based On Neural Network

With the advent of the era of big data, the traditional telecom operators and Internet companies faced with the dual pressures of industry peers because of rapidly development of mobile Internet. The transformation of traditional telecom operators made an urgent request to the data-driven operator transformation from traditional telecom operators to enhance the competitiveness of enterprises, data mining technology will have a useless. Telecommunications industry is no lack of data mining algorithms use a combination of different telecommunications data to solve the customer’s precise marketing, customer behavior analysis, developing packages, churn prediction and telephone fraud, and other issues.No matter for what kind business, customers are the most important resource for company. How to retain the valuable and high-value customers is the eternal goal of enterprise. In real life, the loss of customers were without warning and no rules to follow, only to realize user Erosion Predication, early detection of users likely to churn and take proper retention measures in order to prevent the loss of value of the business.Neural network algorithm is a supervised learning classification algorithm which can predict the customer erosion. And the BP Neural Network is one of the most widely used feedforward neural network, which is a back propagation to update the weights. BP neural network we use to build customer churn prediction model, when large-scale data processing, network training time will be very long, so we use the MapReduce programming framework to parallel algorithms in Hadoop. The map calculate the back-propagation weights for each neuron is connected to the local gradient generated by changing the amount of weight, reduce calculate the average of all samples the amount of change. Then use the batch mode will change the amount of the average of the sample used to update the weights.In this paper, we extracted from the plurality of subsystems of a telecommunications company in the customer information, account information and mobile users for voice, data and value-added services information, through the method of data integration, cleansing, statute, conversion and then generating a training data set. Use BP neural network algorithm to build customer Erosion Predication model of telecommunications users, and the prediction hit rate can reach 82.12%. At the same time, we use MapReduce programming framework make BP neural network parallel in Hadoop platform. Effectively solves the problem of slow speed stand-alone operation, simultaneously ensure the precision and accuracy of models. The experiments show a clear advantage in handling massive amounts of data based on BP neural network algorithm used MapReduce programming framework.