Research On Customer Classification And Churn Prediction In Fitness And Leisure Industry

With the intensification of competition among enterprises,the center of corporate marketing has shifted from products to customers,and customer relationship management has become a core issue for enterprises.Under the premise of limited enterprise resources,accurate customer value classification is an important basis for enterprises to optimize the allocation of marketing resources.As a powerful resource for enterprise competition,customer retention is an important part of enterprise cost control,and the identification of potential lost customers is the primary task of customer retention.Therefore,combining the differences in customer value to accurately predict the loss of customers with different values,so as to concentrate limited resources on the retention of high-value customers,in order to achieve the goal of maximizing corporate profits,is a very hot research topic at present.This paper takes customers in the fitness and leisure industry as the research object,and explores the problem of value classification of customers in this industry and the problem of churn prediction based on customer classification.Firstly,considering the impact of customer classification on predicting customer churn,this paper improves the original customer value recognition model RFM based on the analysis of the impact of customer historical consumption behavior on customer value.It replaces the original last consumption proximity index（R）in the RFM model with the average proximity of the last 3 consumptions,and introduces the customer’s average consumption time indicator D to construct a new customer value recognition model RFMD.In order to solve the shortcomings of the K-means algorithm in determining the initial clustering center,this paper proposes to combine two-step clustering and K-means clustering,and designs an improved two-step clustering algorithm to effectively classify customers based on the RFMD model.Next,this paper starts from industries where transaction types are charged on time and consumption time depends on customers.Based on customers’ historical consumption time and value data,this paper establishes a data-driven customer churn prediction framework containing 26 variables.In the model training and prediction stage,this paper proposes to integrate Lasso regression（L1 norm）and logistic regression with L2 norm to give full play to the feature selection function of L1 regular term,and the advantages of L1 and L2 in reducing the risk of model overfitting And logistic regression can provide strong explanatory and robust results on binary classification problems.First,the Lasso regression feature selection method is used to screen the characteristics of different value customer groups based on the feature framework,and the variables with high predictive ability are retained respectively.Then,a logistic regression algorithm with L2 regular term was used to construct the churn prediction model of different value customers to obtain the churn situation and driving factors of different value customer groups.This paper applies the improved model and algorithm to practical application scenarios,and uses a data set of about 380,000 consumptions of 50,672 customers in a Shanghai stadium to perform model verification and comparative analysis of results,which has realized the effectiveness of customers in the fitness and leisure industry.Divided and predicted the loss of customers of different values in the industry.The research found that:（1）Based on the RFMD customer value identification model,the stadium customers are effectively divided into three groups: important value group,general value group,and low-value group.（2）The churn prediction model built on the basis of the churn prediction customer feature framework based on the Lasso regression and logistic regression fusion algorithm has better performance,and the customer classification obtains better prediction results than before the classification.Among them,the Accuracy and Precision of the important value customers Compared with F1-score,the overall increase was approximately 6%,5%,and 3% respectively,and its true case rate reached 100%,and no churn customers were misjudged.（3）The churn prediction characteristics of different valued customer groups are not completely the same,that is,the main churn driving factors and influence levels are different.