A machine learning approach to predicting customer churn in the banking industry using CRISP-DM and KNIME analytics

Attracting new customers is more expensive than maintaining the existing ones. One way of preventing customers from churning is to develop techniques for predicting their likelihood to churn. This study aims to forecast the customer churn of a multinational bank by using one-month period data, from July 31, 2022, to August 29, 2022. We employed CRISPM-DM, in conjunction with KNIME Analytics, to build several predictive models, i.e., decision tree, random forest, logistic regression, artificial neural networks, support vector machine, and ensemble models to predict customer churn. The results show that the random forest model has the highest performance in accurately predicting the churn of bank clients by its high overall accuracy of 78.91% and AUC score of 85.3%. The decision tree model, with an overall accuracy of 71.55%, and the logistic regression model, with an overall accuracy of 71.3%, are the least-performing predictive models. The findings also show that the customer's historical record with the bank (product_number), credit score, and age have the highest predictive power for customer churn. This study offers valuable insights for financial institutions. Using reliable predictive models, banks may identify potential clients likely to switch to other financial institutions. This identification would allow banks to design innovative marketing strategies and powerful customer relationship management to prevent consumers from churning.