Predicting Customer Churn in E-grocery Service using Machine Learning Model

This work is devoted to studying the problem of customer churn in the field of online food delivery (e-grocery) and developing practical recommendations for customer retention based on a predictive machine learning model. The purpose of the study is to create a model for predicting the probability of churn of e-grocery service customers and to propose a set of preventive measures and loyalty programs to reduce churn and increase customer satisfaction. The work analyzes the features and trends of the e-grocery market, examines the key factors and consequences of customer outflow, and also studies existing approaches to forecasting outflow in industry practice and scientific research. Based on real data on the behavior and characteristics of e-grocery service clients, a churn forecasting model was developed using the CatBoost gradient boosting algorithm. The quality of the model was assessed using various metrics (ROC-AUC, Precision, Recall, F1) and key factors influencing churn in different customer cohorts were identified using model interpretation methods (SHAP). Based on the results of the analysis, a set of measures was proposed to improve the quality of service, personalize marketing communications through CRM and conduct regular research to identify points of growth and improve customer experience. Detailed recommendations have been developed for the implementation of an outflow forecasting model in the business processes of the e-grocery service. The results of the work have practical significance for improving the efficiency of customer churn management and personalizing marketing strategies in e-grocery, and can also be adapted to other industries and business models where customer retention is a key success factor. Keywords: customer churn, churn prediction, machine learning, gradient boosting, CatBoost, CRM.