Research On Rail Freight Customer Churn Warning Driven By Data

With the rapid development of China’s transportation and the further optimization of the transportation structure,the volume of rail cargo volume has increased significantly during the"13^th Five-Year Plan"period.However,rail freight has always faced greater operating pressure and the risk of customer churn in the competition with other ways of freight.Therefore,under the customer-centric management model,the study of customer churn in the field of rail freight can not only provide a theoretical basis for customer relationship management,but also assist the rail freight department to take corresponding retention measures in time,which can effectively reduce the customer churn,enhance the competitiveness of the rail freight market and increase the market share.This study aims to establish a characteristic engineering index system of rail freight,enrich the research on rail freight customer churn warning,and provide a theoretical basis for subsequent scholars.At the same time,the machine learning methods are applied to the research of customer churn warning in rail freight field,and the model fusion method is adopted to improve the accuracy of model prediction,so as to put forward the corresponding retention strategies and save the marketing costs.First of all,this paper takes the historical order data of real freight customers of C Railway Group Company in China as the research object.Based on the understanding of rail freight data characteristics and customer delivery behaviors,it explores the historical order data,specifically defines the label vector of rail freight customer churn sample,and sets the time window.The RFM model was extended to construct rail freight customer churn warning feature project,and six important feature indexes were obtained through feature correlation analysis and importance ranking:R,F,M,D,K and V are used to construct the characteristic matrix of rail freight customer churn warning model.Second,use Random Forests,GBDT,LightGBM three machine learning methods to establish customer churn warning mode,train the models based on the data labels defined in this paper and the established feature dimensions.The prediction and generalization ability of each model are verified by the test set and the cross-time sample test data set respectively.Next,based on the stacking ensemble learning method,the fusion model is constructed,and the first layer uses random forest,GBDT and LightGBM algorithms as the base classifiers,and the second layer adopts the logistic regression algorithm as the meta classifier.Finally,analyze the prediction results of the fusion model,set up a customer churn warning mechanism and give the customer retention strategies to help the rail freight department to formulate a customer churn warning plan with visual risk status and clear action information.The experimental results of Stacking fusion model show that the AUC value is 0.8461,the accuracy rate is 80.80%,the recall rate is 90.45%,and the F1 Score is 84.27%,which proves that the fusion model has better prediction and generalization ability than the three base classifiers.This study provides a new warning method for customer churn of rail freight,and further helps guide rail freight companies to reduce customer churn rate,thereby reducing operating costs and expanding market share,which has high practical significance and practical application value.