Customer churn prediction for commercial banks using customer-value-weighted machine learning models

Customer churn prediction has become an increasingly important issue in global business, especially in the banking industry, where customer acquisition has become ever more costly in this notoriously competitive business environment. Although many methods have been proposed to solve this issue as a classification problem, there are few studies that consider customer values in the light of attrition analysis. In this paper, we propose a framework to address this, and we quantify customer values with the use of an improved customer value model, examining them from the perspective of their recency, frequency, monetary value and asset level. We take customer values as the basis of misclassification costs that, in turn, direct machine learning predictive models. The returns for banks in this scenario can be maximized, given various cutoffs and some assumptions. This proposed framework may provide commercial banks with useful insights to better formulate marketing strategies for different groups of customers, as well as to analyze attrition in an economic way, rather than as a simple classification problem.