Prediction of Customer Churn: A Boosting and Tree-based Machine Learning Framework for CRM and ERP Systems

Customer churn prediction is the most critical aspect in the business world, where sales are considered an essential factor in growing the business. Identifying the customers who are about to discontinue some subscriptions is crucial to providing them with discounts and personalized offers so that they can return and buy the products. Thus, this research focuses on understanding customer behavior and identifying the customers who can leave. Also, the features that are responsible for predicting customer churn are identified. For this, the open-source dataset is used to implement the ML approach, which requires the initial data cleaning where outliers are detected, missing values are identified, and the dataset is balanced. The balanced dataset is provided as an input to the ML algorithms like LGB, XGB, CatBoost, HGB, and RF. The highest-performing algorithms are LGM, with 96% accuracy, and RF, with 95% accuracy. The most important features identified by the highest performing model LGB are total charges, monthly charges, tenure, gender_male, and Paper_less billing. Hence, it is understood that the male customers are more likely to churn than the female customers. Implementing the proposed framework in real-time can help the business grow by predicting customer churn and providing such customers with the best possible offers to enhance the company's profit.