Research On Customer Churn Prediction Based On Customer Segmentation

With the continuous deepening of the degree of global economic integration,more and more new companies and new products have entered the market,which has intensified the competition among enterprises,and the competition of enterprises is mainly reflected in the competition of customers.At the same time that competition will lead to customer churn,so customer churn is an important issue facing companies.At the same time,for the enterprise,human,material and financial resources are limited,and customer values are different.The rational and differentiated allocation of limited resources is also an issue that enterprises need to solve urgently.Therefore,this article comprehensively considers the problems of customer churn and customer value,and performs customer churn prediction research based on customer segmentation.After carefully studying and analyzing the research results on customer value,customer segmentation and customer churn prediction at home and abroad,the research content,research method and research route of this paper are put forward.The data in this article comes from the 2016 "Church Prediction of Customer Churn" competition,which is the hotel customer churn prediction data provided by Ctrip.com.First,the data is pre-processed and derived variables are constructed through feature engineering.The size of the processed data changes from 689945×51 to 689945×83.Second,establish a customer segmentation model based on customer value.Combined with the characteristics of the data selected in this paper,based on the four aspects of income value,growth value,customer loyalty,and word of mouth,a customer lifetime value evaluation system with 22 indicators was constructed.The K-Means clustering method was used to segment customers.Divide 3 different customer groups,and use principal component analysis to comprehensively analyze the load plus and minus and the size of each principal component,and select the principal component that represents value,so as to calculate each type of customer group on the principal component.Average score,and then quantitatively divide customers into high-value,medium-value,and low-value customers based on the average score.Finally,for the high-value,medium-value,and low-value customer groups obtained by customer segmentation,a customer churn prediction model is constructed.The training models selected in this paper based on machine learning methods are random forest,XGBoost,and LightGBM.The models are evaluated by Accuracy,Precision,Recall,F1-score,and AUC values.In order to further verify the prediction effect of the model after customer segmentation,compare and analyze the performance of the model on the evaluation indicators before and after the customer segmentation,and finally found that after the customer segmentation,the overall evaluation indicators that consider the accuracy and recall rate are all The improvement has provided ideas and directions for customer churn prediction and retention to a certain extent.