Simulation And Research On The Prediction Model Of Telecom Customer Churn

The reorganization of the telecommunications industry, get rid of a dominant monopoly market, into a situation of tripartite confrontation of the competitive situation. The information age, the competition is the key to success or failure to win by the information practice conversion rate, which is a new information processing tools, application. Due to the research results of this paper, the first line staff for Telecom enterprises. Therefore, we select the decision tree and neural network based on two basic algorithms to set up the model, get rid of annoying manual complex operation, simply call the target data can be automatically calculated results. Easy to understand, simple and efficient.In the decision tree classification, but limited to small scale data. Neural network in large-scale data processing, but the sample dimension limits. This will be the two comprehensive use by decision tree neural network reduction, scale of decision tree based on neural network, complement each other, each in his element. In the current situation of limited resources, this paper selects the customer value evaluation model to screen the potential loss customers for two times, select the high value customers to maintain. In order to ensure the cost of recycling, and future earnings. Model the overall operation procedure is as follows: firstly, using the decision tree model selected attribute index loss characteristics of the sample data, extract the characteristic data of the sample data; then, using neural network model to screen the potential customer loss characteristics in the data; finally, using the customer value evaluation model to determine the high value customer potential customer churn the loss of ownership as the target.After the completion of the preliminary model, the marketing center of a telecommunications company from my city during the period from December 2014 to May 2015, the sample data of 6000 samples of the 15 customer attribute index for a total of 540 thousand points, the model of the test run. During the testing process, the model parameters are adjusted repeatedly and repeatedly, so that the operating results of the model in the three dimensions of coverage rate, hit rate and correct rate reach the set target. Into the actual production, the effect of good response.