Segmenting Credit Card Behavior using Agglomerative Clustering

This research uses agglomerative hierarchical clustering to classify credit card users into behaviorally different segments, facilitating data-driven marketing. Examining a large database of financial information with ward, complete, and average linkage methods, we identified two distinct behavioral clusters: specifically, k equal to two, an engaged segment comprising two thousand high-activity customers characterized by one-time purchases averaging one point two five, transactions averaging zero point eight four, high credit utilization, and sound financial health reflected in balance and credit limit measures, and a conservative majority of six thousand eight hundred inactive customers with low activity and negative measure values. Ward linkage was superior to the other methods, with a calinski-harabasz index of two thousand four hundred twenty-eight point one zero nine and a silhouette measure of zero point eight nine six, while average linkage was verified with clear cluster boundaries in principal component analysis and t-distributed stochastic neighbor embedding plots. Feature importance analysis identified balance, contributing forty-eight percent, as the predominant driver, followed by purchases at thirty percent and payments at twenty-two percent, with tenure and cash advances demonstrating no influence. These results allow institutions to make precise adjustments to services, optimize marketing, and enforce customer retention through precise behavioral segmenting. Future studies will include customer input to enhance cluster validity and dynamic adaptability. The binary segmentation provides actionable insights for financial institutions, enabling targeted marketing strategies for high-engagement customers and activation campaigns for conservative users.