Predicting Bank Customer Churn: An XGBoost Approach to Enhancing Customer Retention

This study applies the XGBoost classifier to analyze the propensity of the banking’s customer churn, by analyzing the dataset containing the basic demographics, financial records, and churn label. Some of the features captured in the dataset include; Surname, Row Number, CustomerID, Credit Score, Gender, Age, Tenure, Account Balance, Credit Card, Number of Products, Geometric Mean, Active Member No., Projected Salary, Exited where Exited = 1, the customer has left the bank. The purpose of present research is to apply the predictive analytics in order to understand what underlying factors serve as indicators of customer attrition and in turn facilitate the better strategies of the relationships’ maintenance. The model scores an average accuracy of 85%, based on 2000 samples, relatively well for predicting non-churning customers (Class 0) with an accuracy of 0.90, recall of 0.93, and an F1 of 0.91. Nevertheless, low accuracy (0,65), recall (0,56), and F1-score (0,60) were received for the churned customers (class 1), which means a high level of false negatives. These results also reveal that the task of identifying churned customers is indeed difficult especially when working with imbalanced data sets. Thus, the macro average performance shows that the model is not very good at generalizing, especially in the case of the minority class. This paper suggests that there is a need for further model improvement including hyperparameters tuning and implementing ways to address the imbalance of class problem to enhance sensitivity in detecting churned customers. Improving the model’s accuracy at identifying ‘at risk’ customers will go a long way to helping refining customer retention strategies for the banking industry, thus helping cut attrition rates and increase overall satisfaction.