SEGMENTING BANK CUSTOMERS TO EXPLORE PRODUCT ENGAGEMENT OPPORTUNITIES : A Comparative Study Combining Clustering and Predictive Modeling within Banking

Sammanfattning

Understanding patterns in customer behavior is a central challenge in data-driven banking. This study explores how customer segmentation and predictive modeling can be combined to uncover engagement opportunities for savings products within a large mortgage customer base. Using a dataset of bank customers, we first apply clustering techniques to identify distinct customer groups based on behavioral, demographic, and financial features. Two complementary clustering approaches are used: K-Prototypes for distance-based segmentation of mixed-type data and HDBSCAN for discovering density-based clusters after dimensionality reduction with UMAP. The quality of the clusters is evaluated using internal metrics. Following the segmentation, we develop predictive models to assess the likelihood that a customer adopts an additional product—in this case, a savings account. We compare logistic regression with XGBoost in terms of classification performance, using precision-recall AUC as the primary evaluation metric due to class imbalance. The results demonstrate that segmentation provides valuable structure for understanding customer behavior, and that combining clustering with prediction can enhance the bank’s ability to find potential opportunities for further engagement. The analysis reveals that certain customer segments have a significantly higher or lower probability of holding a savings account, enabling more targeted and effective engagement strategies. This integrated approach offers a practical and scalable framework for improving customer understanding and identifying areas with untapped engagement potential in data-rich financial settings.