The Study Of Mobile Client Phone Churn Tend Prediction Model Based On Support Vector Machines

1. IntroductionCustomer churn is a rational consumer choice. Its occurrence has a very clear causal relationship. This causal relationship is often reflected in the records of past consumption, so consumption of the past records of customers research to analyze customer behavior essential for the loss of a process. Customer churn behavior in the previous case study of consumer behavior is rare, but the negative effects on the loss of enterprise customers is quite difficult problem to enterprises. Through the history data of previous analysis, customer churn analysis is to find off-grid customers characteristics, and timely to take appropriate measures, and to reduce customer churn happening. It is extremely significant for this enterprises to reduce operational costs and improve business performance. Therefore, the research to explore new customers evaluation is still in the ascendant [1]. Moreover, since the competition between mobile operators increasingly fierce, from the subjective increase in mobile phone customers likely to take the initiative away from the net. This takes difficulty in assessment of mobile phone customers unpayment, and objectively requires an assessment of the system from the algorithm that is continuously updated and improved. Therefore, the forecasting system algorithm has a very important research value and strategic significance.According to statistical learning theory, and structural risk minimization principle of the initiative, support Vector Machine (Support Vector Machine, SVM) is proposed by Vapnik and his team. SVM is a new data mining method able to handle very successful return of many problems in pattern recognition problems and SVM can also be extended in the areas of forecasting and evaluation, widely used in statistical classification and regression.SVM can maximize the generalization ability of learning machines, even if the limited data set from the discriminant function of the moral right of independent test set, which can still get a smaller margin of error. In addition, the support vector machine is a convex quadratic optimization problem, and to find the extremal solution is a global optimal solution. These characteristics make SVM become an excellent opportunity for data on the machine learning algorithm. LibSVM is a simple, easy to use, fast, efficient SVM pattern recognition and regression software program, which not only provides a windows-based operating system compiled executable file, but also provides information on the software source code to facilitate the improvement, modification as well as in other operating systems applications. The software SVM involves parameter adjustment is relatively small, providing a number of default parameters, using these default parameters can be solved many problems, and provides cross-validation (Cross Validation) feature. Using the software program can solve C-SVM classification, v-SVM, e-SVM regression and v-SVM regression and other issues. Using support vector machine therefore identified as a predictor of core algorithm is a good choice.In summary, the main task of this topic and purpose of the algorithm is to use LibSVM tendency to predict the establishment of mobile customer churn classification model, through LibSVM parameter optimization in order to establish the best predictive models to improve the efficiency of identifying customer churn prediction, and accuracy.2. Research ContentThis algorithm is based on the theoretical model predictions, to mobile phone users tend to off-grid forecast for research purposes, describing the data collected from the customer forecasts for the entire process. To support vector machine algorithm based on the basic theory, research using Libsvm off-grid algorithm is used to mobile phone users tend to predict the application. Main research contents are as follows:(1)The basic idea of support vector machines, working principle, evaluate the advantages and disadvantages of various algorithms, and application at home and abroad, linear separability and linear inseparability from the two side briefed Le SVM's, and constitute an important core function module. By the polynomial function, radial basis function and Sigmoid function, these three commonly used kernel function describes the algorithm kernel features and functions.(2) Study of Libsvm model, deep discussion of the model of the software features and mathematical theory and the use of LIBSVM key technologies; by C Support Vector Classification (Binary) and v Support Vector Classification (Binary) detailly describes the use of LibSVM and the SVM main formula, from the C-SVM decomposition algorithm, working set selection and stop the cycle of standard, decomposition method, analytical method, and calculate the quadratic programming problem . From the algorithm, the set of compression, multiple classification, unbalanced data set, calculation method and parameters to find the basic principles of Excellence, for the prediction model was established as the basis and theoretical basis of the algorithm.(3) To discuss characteristics of mobile phone customers away from the net, to look for quick and simple feature extraction method of off-grid customers, with the telecommunications industry for China's own characteristics, to combine with China's basic national conditions and the actual operation of the feasibility of the business comes up with the collection of the needed data field. According to the proposed program, the related data collection shall be implemented. Followed by 2 times the data filtering and normalization, the data clearly predicted in advance the process to be considered and resolved. Cell phone-based customer churn LibSVM tendency prediction algorithm develops a forecasting model of data processing, as well as that the prediction algorithm has laid a data base.(4) Changes in support vector machines using the initial weights and threshold parameter optimization manner, through C and parameters by grid search to change the structure prediction accuracy based on the optimal value. In the paper the optimization parameters determined by when the C = 8129.0 and = 0.03125 when the model prediction accuracy on 98.324%, reaching the peak.(5) Based on mobile customer churn Libsvm tendency prediction model, and simulated tests, the accuracy rates were 97.35%, algorithm accuracy shall be verified. Tests show that in the present study data collection, using LIBSVM model building, in the parameter C = 8129.0, = 0.03125, the forecast accuracy is optimal, which were 97.35%, fully explaining that it is successful tendency predict to use study data collection method and to combine mobile customer churn LIBSVM algorithm, and get a higher tendency to predict customer churn rate of speed and accuracy, indicating that a prediction algorithm based on LibSVM tendency to predict customer churn in mobile phone has certain advantages.3. Conclusion:Tests show that in the present study data collection, using LIBSVM model building, in the parameter C = 8129.0, = 0.03125, the forecast accuracy is optimal, which recognition accuracy is 98.324%, fully explaining that it is successful tendency predict to use study data collection method and to combine mobile customer churn LIBSVM algorithm, and get a higher tendency to predict customer churn rate of speed and accuracy, which is designed to illustrate features of the phone customer data extraction programs and support vector machine prediction algorithm after mobile phone customers away from the net prediction tend to the timeliness and accuracy. The accuracy rate can reach 97.35%.However, by selecting the appropriate forecasting techniques, the establishment of mobile phone customers away from the net tendency prediction model to predict the operation, the prediction value or use other methods, the prediction value of the enterprise, is still just an initial value. This is because developments from the past to the present can not be deemed as future changes, but also refer to the current various possibilities, as well as new trends and developments, and comprehensively analyze, compare, judge, assess and finally adjust and amend the predict results.