The customer churn prediction system for banking services on machine learning platform

Abstract

<p><span style="font-weight: 400;">The increasing integration of artificial intelligence and the Internet has revolutionized data analysis and processing, particularly in customer behavior prediction. However, in the banking sector, machine learning techniques have not been widely adopted for churn analysis, with existing approaches often relying on outdated, static demographic information. This study proposes a machine learning-based model for predicting customer churn using historical transactional data from DEBIT accounts. The dataset was collected through formal agreements with three commercial banks in southern Vietnam, utilizing the CASA transaction information system. Data collection and usage complied with strict confidentiality protocols, ensuring legal validity and data reliability. After a comprehensive data preprocessing pipeline, several machine learning algorithms were evaluated, with Random Forest achieving the highest performance with 96% accuracy. The proposed model has been integrated into the banks’ Customer Relationship Management (CRM) systems to assist in early identification of potential churners and to inform retention strategies. The findings contribute theoretically by demonstrating the importance of dynamic behavioral indicators over static demographics, and practically by offering a scalable solution to improve customer retention in the banking sector. </span></p>