Research On Analysis And Prediction Of Mobile Consumer Churn Based On HMM

With the continuous development of the communications industry and the further deepening of structural reform,the domestic telecommunications industry within the major operators intensified competition.The competition among the internal operators of the communication industry is mainly the contention of consumers.The consumer base directly affects the operator's revenue and profits.Operators in order to obtain more consumer resources and occupy a greater market share,tend to take price war and advertising,such as virtual operators through low price war to attract new consumers,leading to the traditional three basic operators have to reduce the tariff to retain consumers,prompting the communications industry to speed down the fee.The paper designs a probabilistic model of consumer churn based on the communication data of mobile operators.First of all,dealing with the features of the original communication data,by analyzing the features of communication data to extract the data hidden features,and build data features by data discretization,and then select features of data.The three-step purpose of this paper is to make the processed data better reflect the willingness of the consumer churn.Then,the data model is modeled based on Hidden Markov Model.The probability model is constructed by hidden states and observation States in the timing of the process of probabilistic reasoning,and propose more effective probability forecasting model based on Naive-Bayes Markov chain model.Finally,based on the big data volume of mobile call records,it is impossible for any single high-performance computer to carry out high-efficiency prediction.In this paper,all the contents of the design were based on MapReduce programming model of parallel design.In this paper,consumer churn prediction model based on mobile communication data is designed and implemented based on MapReduce,and the probabilistic reasoning model is validated and analyzed by using the real call records data set of Liaoning Branch of China Mobile.The experimental results show that the proposed model has good preformances in the real data set.And it is feasible and effective.