Predicting Bank Term Deposit Subscription Using Machine Learning Models

Abstract

Banks are increasingly relying on data-driven target positioning to transfer potential customers into product buyers. By effectively identifying customer characteristics, one can reduce their marketing costs. However, converting purchase propensity models into measurable data remains challenging. This article integrates the advancements in tabular data modeling and marketing analysis to bridge this gap. It first provides an overview of the banking context and dataset characteristics, and then compares the propensity evaluation of basic linear models with ensemble methods, emphasizing the role of interpretability in compliance and on this basis, multiple models were established and evaluated, then the most suitable model was selected based on the data sets. Besides, this article discusses evaluations beyond accuracy: by using ROC/P&R analysis to link threshold selection with operational trade-offs. To support targeted actions, customer segments are defined, groups that are price-sensitive are identified for tailored interventions, so that the customer value are optimized. Thus, this article provides a practical perspective for transfer predictive data into profitable and auditable banking activities.