| In today’s era of intelligent e-commerce,the business competition model has also quietly changed,and personalized service to customers has become a decisive factor for the survival and development of enterprises.Because of the high cost of customer churn,it is necessary to consider differentiated services for existing customers to ensure the reasonable allocation of enterprise resources,so the content of customer churn prediction has become an important part of customer relationship management.In this paper,we analyze the behavioral data of e-commerce customer data by studying the literature related to e-commerce customer churn prediction,and find that there are characteristics such as imbalance in the data,carry out e-commerce customer churn analysis,and propose an e-commerce churn prediction model based on customer segmentation.The main work of this paper is as follows.First,Due to the large amount of e-commerce customer information,customer consumption behavior records are selected to establish a customer value RFM model,and a segmentation model is built for enterprise customers based on customer value,the optimal number of clusters is determined by the contour coefficient to be 4,and KMeans clustering is used to classify into four types of customer groups,and later the weights of each index of the RFM model are objectively determined by the entropy weight method to determine the value levels of various customer groups.Second,to address the imbalance on the churn characteristics in e-commerce customer data,a boundary-based hybrid sampling imbalance data processing algorithm(UBMS)is proposed,which introduces the concept of supporting k-outlier degree threshold and boundary factor,and divides the boundary by comparing the two.The new samples are synthesized by SMOTE algorithm for the minority classes at the boundary,and the improved clustering undersampling is performed for the majority classes not at the boundary,and the sample center position of each cluster is derived based on the dimensionally weighted Euclidean distance between the calculated samples,and the majority class samples with stronger information features are selected according to the cluster center neighborhood range to delete the invalid information points in the e-commerce data so that the overall state of equilibrium is achieved.The UBMS algorithm can effectively improve the recognition performance of minority class samples.Third,based on the above,the overall framework of e-commerce customer churn prediction is proposed,and the optimized Light GBM algorithm is used as the classifier of the prediction model.The overall data are compared with the data after customer segmentation and different classifier models,and the analysis results can be obtained:customer segmentation is beneficial for better e-commerce merchant churn,and the optimized Light GBM algorithm is more suitable as the prediction model in this paper. |
