Research On The Method Of Early Warning Of Customer Churn And Return To Factory For Automobile Dealers

With the rapid development of China’s auto market,the buyers are constantly turning into after-sales customers in the market,and the auto dealers have rapidly accumulated a large number of customer resources.But now the car market is in the doldrums.The contradiction between production capacity of car enterprises and market demand is becoming more and more serious.Customers are also abandoning car dealerships as they have more options.With fewer people buying cars and fewer existing customers,the total number of car dealers’ customers continues to decrease.The car market in our country tends to be saturated,the overall profit of the industry declines,and the era of competitive market is coming.Customers are the fundamental resources and the biggest assets of an enterprise.The loss of customers also means the loss of enterprise assets.Nowadays,the way of customers choosing products and services presents liberalization and diversification,and customer loyalty is getting lower and lower.Customer churn has become a key issue affecting the future development of auto dealers.On the other hand,the rapid,efficient and accurate identification of customer risk is a key measure of customer relationship management.Customer churn warning method is used to classify customer risk,which can prevent or avoid customer churn and improve the service level and competitiveness of auto dealers.With the rapid development of information technology,more and more enterprises store transaction information in databases,which provides data resources and platforms for customer churn risk assessment.And machine learning algorithm research provides a lot of theoretical methods to improve the accuracy of customer churn model.This paper takes the data of Case A of "The Fourth MAS Case Competition" as the empirical research object,combs the process of customer churn risk assessment of auto dealers,and compares the effect of customer churn risk assessment model.This paper builds Wo E-Logistic model to solve the problem of customer churn risk assessment.The concepts of Weight of Evidence and Information Value are often used in the study of risk assessment problems.Then,based on the three data mining algorithms,Random Forest,Ada Boost and XGBoost are applied to the processed data to further improve the accuracy of the model.Finally,the customer churn risk score is graded by Information Value.For the prediction model of return time,firstly,the return time interval was constructed as the target variable,and the logistic regression model and XGBoost algorithm were used to predict the return time.According to the predicted return time,high-risk customers who predicted return but did not return to the factory were identified,and high-risk customers were successfully monitored.Finally,according to the risk level,the characteristics of customers who have not returned to the factory are analyzed,so as to provide differentiated services for customers with different risk levels and achieve accurate service for customers.After comprehensive over-sampling treatment of sample imbalance,the results show that the recall rate of Logistic model is only 16%,while the recall rate of Wo ELogistic model is 79%.WOE-Logistic model is more accurate in the identification of lost customers.Comparing the three machine learning algorithms,it is found that XGBoost algorithm has the best optimization effect,which increases the recall rate to90%.Through the customer risk rating table,the scores of different value ranges of each variable are calculated,and it is found that car age,sales price and loan term have a great impact on customer loss.Finally,the customer churn risk score is divided into five levels according to Information Value.Moreover,combined with the return forecast results,29,846 customers who did not return were identified,among which56.51% of the customers with Level 5 loss risk were accurately monitored.