Deep Embedding Clustering with Adaptive Feature Selection for Banking Customer Segmentation

Abstract

Customer segmentation is critical for financial institutions to optimize marketing strategies in the competitive credit card market. Traditional clustering methods such as K-Means often fail to capture complex behavioral patterns and lack interpretability for business decision-making. This paper proposes a deep embedding clustering framework with adaptive feature selection for banking customer segmentation. The method integrates a stacked autoencoder to learn low-dimensional customer behavior representations while preserving interpretability of business-critical features. A mutual information-based feature importance mechanism automatically identifies discriminative behavioral indicators. Furthermore, a business-constrained clustering optimization approach ensures segmentation results align with marketing objectives. The framework will be evaluated on a real-world dataset of credit card customers from a major U.S. financial institution. This research aims to advance AI-driven precision marketing, supporting banks in optimizing resource allocation and enhancing customer lifetime value.