Perfilamiento de clientes bancarios con un enfoque estratégico para la personalización de productos o servicios financieros

Abstract

Abstract: This paper develops a bank customer profiling model designed to personalize term deposit offers, using the Bank Marketing dataset (41,188 records; 2008 2010). Following the CRISP-DM methodology, preprocessing was performed to preserve the behavioral load of missing non-random data (MNAR). This included variable recategorization, outlier removal using Winsorization at the 95th percentile, and spatial standardization (StandardScaler and One-Hot coding). Two scenarios were constructed to prevent predictive data leakage. One is a pre-campaign scenario, designed for prevention, and the other is a post-campaign scenario, used to understand customer behavior. Algorithmic selection was based on internal validation metrics (silhouette and Davies-Bouldin), adopting K=8 clusters for both phases due to their maximum cohesion and interpretability. The partitioning patterns differed across the commercial market, with the elite segment achieving over 63.7% subscription, while the low-performing groups reached only 3.7%. Validation using Gaussian Mix Models (GMM) confirmed structural stability, and the DBSCAN algorithm identified a topologically atypical microsegment (3.76% of the sample) with a 50.9% conversion rate. Taken together, the results provide an actionable framework for commercial prioritization and targeting high-value niches.