The Research For Customer Churn Prediction Model Based On Lasso And RBF

The increasingly saturated market has made it more difficult for companies to expand their market share,and the growth of company customers consumes more costs for enterprises.Therefore,how to stabilize the existing customers has become the company's concern.The loss of existing old customers not only brings huge economic losses to the company,but also weakens the company's social influence.Therefore,predicting churn for corporate customers in advance and proposing targeted marketing strategies will become the main players in retaining customers and maintaining stable business development.Nowadays,big data analysis technology is widely used,and it is inevitable for companies to use their data assets to guide business decisions.In the process of customer churn prediction,by modeling and analyzing complex big data scenarios such as high feature dimensions,dynamic changes in features,and uneven data,the customer churn tendency is accurately predicted to provide an effective basis for corporate marketing strategies.The thesis proposes an RBF optimization model based on Lasso regression,combined with the customer life cycle,establishes a customer churn prediction model,and verifies the model based on a public data set of a bank and a desensitization data set of a local state and telecommunications company.The main work of the thesis includes:1.Aiming at the problems of severe imbalance between churn and non-churn customer data,high dimensionality of customer data,and dynamic changes in eigenvalues in customer data information,an RBF optimization model L-RBF based on Lasso regression is proposed.The model uses a hybrid sampling method to balance the data set,thereby improving the target recognition rate.In order to improve the flexibility,generality and prediction effect of the model,an RBF neural network algorithm parameter model is further constructed.Through experimental comparison and analysis with algorithms such as Logistic regression,RBF,and Boosting-L,the results show that the L-RBF model has a higher recall rate and better predictive classification ability.In addition,the application verification of the L-RBF model was performed,and the features extracted from the model were analyzed and explained in order to make preliminary suggestions on the problem of customer churn.2.Due to the different requirements of telecommunication customers for enterprise products in different periods,the spatial characteristics of customer samples and their attribute values are also different.To this end,the concept of customer life cycle is introduced,and a telecommunication customer churn prediction model based on the entire life cycle is proposed.The model divides the data set into multiple subsets according to different stages of the life cycle,and apply the L-RBF model to obtain the full life cycle feature attribute set.At the same time,the trend analysis and association analysis are performed on the feature attributes of the full life cycle to facilitate Phased targeted marketing.Through experimental comparison with the non-life-cycle data set partitioning,the results show that the prediction model based on the whole life-cycle has better results in the areas of target recognition and association feature acquisition.The research work shows that the telecommunication customer churn prediction model based on the L-RBF optimization model and the entire life cycle can not only effectively predict customers with a churn tendency,but also reduce the impact of customer characteristics under different business requirements on the prediction results.Thereby,the accuracy of the model can be effectively improved,which has good theoretical research and practical significance.