An Improved Machine Learning Based Customer Churn Prediction for Insight and Recommendation in E-commerce

Abstract

Since keeping existing customers costs far less in e-commerce, recruiting new customers is no longer a wise approach. Therefore, businesses are increasingly putting greater emphasis on lowering their customer churn rate due to the level of competition present in the business-to-consumer (B2C) e-commerce arena and the significant investments necessary to recruit new customers. Large volumes of data about their current customers’ transactions, searches, frequency of purchases, etc. are typically held by e-commerce businesses. Artificial intelligence (AI) can be used to evaluate customer behavior and predict potential customer attrition, allowing for the adoption of targeted marketing techniques to keep them as customers. In this paper a customer churn forecasting framework has been developed using the best classifier for insight and recommendation in order to improve the accuracy of forecasts of customers who would churn and make it simpler to identify non-churn consumers. There are five components in the framework, including exploratory data analysis (EDA), data preprocessing, model tuning, comparison among different models after model tuning, insight and recommendation. Experimental results shows that the proposed method can predict customer churn with high accuracy.Accuracy and F1- score are used for model evaluation.According to experimental analysis, CatBoost performed the best in Dataset, with 100% accuracy and 100% F1-score. After selecting the best classifier, the recursive feature elimination (RFE) was applied to find the rank of feature for insight and recommendation so that the paper fills a research gap and contributes to the existing literature in the area of developing a customer churn prediction method.