Customer Value Analysis And Loss Warning Based On Data Mining

Nowadays,with the rapid development of economy,people know better how to protect their rights and interests.They will weigh the pros and cons and choose the most beneficial thing for themselves.The development trend of the market economy is ultimately determined by people.From the survival of an enterprise to the development of the whole industry,if an enterprise wants to survive in the rapidly changing economic situation,in addition to developing and researching its own products,it also need to invest a lot of energy in customer relationship management.Especially in the mobile communication industry,the cake size is fixed and the user groups are defined.If enterprises want to gain greater profits,they need to explore and analyze customers,develop marketing strategies that make them move,and then consolidate and develop customers.This article mainly carries out data mining on customer value analysis and churn prediction of mobile communication enterprise users.In terms of customer value analysis,two unsupervised learning clustering algorithms,K-Means clustering and Gaussian Mixture Model clustering,are adopted because the user data has no category label in advance.Calinski-Harabaz Index was selected for measurement when comparing the clustering effect of the two clustering algorithms.Finally,the clustering results of K-Means clustering algorithm were selected,and the customer value was divided into three levels according to the clustering results.The corresponding marketing strategies are given for customers of different value levels,which are focused on maintaining customers,important developing customers and general value customers.On the customer churn prediction,because the number of churn users and non-churn users is too different,so the subsampling process in random resampling is carried out for unbalanced data set firstly.Subsequently,the CART decision tree,random forest and logistic regression algorithm of the classification algorithm were used in the training process of the model,and after evaluating the model with the evaluation indicators of the binary classification model,the model trained by the random forest algorithm was selected for model optimization,and the parameters were tuned by the way of Grid Search.A model that predicted the recall rate of lost users to be more than 80% was built.Finally,corresponding countermeasures were given according to the main characteristics of lost users.