EVALUATING MACHINE LEARNING MODELS FOR OPTIMAL CUSTOMER SEGMENTATION IN BANKING: A COMPARATIVE STUDY

Аннотация

INTRODUCTION In the contemporary banking landscape, Machine learning algorithms, such as K-Means, understanding customer behavior and segmenting Hierarchical Clustering, Gaussian Mixture Models customers effectively is crucial for developing (GMM), DBSCAN, and Spectral Clustering, have targeted marketing strategies, enhancing shown promise in clustering and segmenting customer engagement, and optimizing service customers across various industries. In banking, delivery. With increasing competition and rapidly these models facilitate the identification of evolving consumer expectations, banks are customer segments with distinct behaviors, leveraging advanced machine learning algorithms preferences, and transaction patterns, which in to segment customers more efficiently and turn supports personalized marketing campaigns, accurately. Effective customer segmentation risk management, and customer relationship enables banks to tailor services, offer personalized management (CRM) strategies (Bolton & Drew, product recommendations, and implement 1991; Gupta & Harris, 2009). strategies that drive customer loyalty, retention, Despite the advantages of machine learning and profitability. models, selecting the most effective algorithm for The shift towards digital banking, coupled with the customer segmentation in the banking sector availability of large-scale transactional and remains a challenge. Each algorithm has its engagement data, presents an opportunity to strengths and weaknesses, and their performance employ machine learning models for customer can vary significantly depending on the dataset segmentation. Traditional segmentation methods, characteristics and business objectives (Everitt et such as demographic segmentation, often fall short al., 2011). For example, while K-Means offers in capturing the complex patterns and behaviors speed and scalability, it assumes spherical clusters, exhibited by customers in the banking sector. which may not always reflect the reality of Instead, machine learning techniques, with their customer interactions (MacQueen, 1967). ability to handle large datasets and uncover hidden Similarly, Gaussian Mixture Models (GMM) patterns, offer a more sophisticated approach to provide flexibility but are computationally segmentation (Smith, 2003; Kumar & Shah, 2006). intensive (Dempster et al., 1977). https://www.theamericanjournals.com/index.php/tajet 69 THE USA JOURNALS THE AMERICAN JOURNAL OF ENGINEERING AND TECHNOLOGY (ISSN – 2689-0984) VOLUME 06 ISSUE12 Existing research has explored the application of LITERATURE REVIEW machine learning techniques in customer Customer Segmentation in Banking: A segmentation, but there is still a lack of consensus Theoretical Background on the most suitable models for large-scale Customer segmentation has long been a strategic banking datasets (Han, Kamber, & Pei, 2011). priority for banks seeking to improve customer Previous studies have primarily focused on relationships, increase profitability, and reduce demographic and transactional data, often risks (Kotler & Keller, 2012). The concept of overlooking engagement metrics and customer segmentation involves dividing customers into interactions across multiple touchpoints (Wedel & distinct groups based on specific criteria, such as Kamakura, 2000). Additionally, comparative demographics, transaction behaviors, or studies that evaluate multiple clustering engagement patterns (Bolton & Drew, 1991). algorithms on large and dynamic banking datasets Historically, segmentation in banking has relied on remain limited. demographic and behavioral attributes, including Therefore, this study aims to conduct a age, income, account balance, and transaction comparative analysis of several machine learning frequency (Smith, 2003). However, these models, including K-Means, Gaussian Mixture traditional methods often fail to capture the Models, Hierarchical Clustering, DBSCAN, and nuances of customer interactions and preferences Spectral Clustering, to determine the most in the digital age. effective approach for customer segmentation in Recent advancements in machine learning offer the banking sector. By leveraging a comprehensive new opportunities for more dynamic and accurate dataset that includes transactional, demographic, customer segmentation. Machine learning and engagement attributes, this research seeks to algorithms can process vast amounts of data, identify the model that offers superior identify patterns, and segment customers based on segmentation accuracy, interpretability, and complex interactions that traditional methods scalability. The study further aims to provide might miss (Han et al., 2011). Clustering actionable insights into how banks can leverage algorithms, a subset of unsupervised machine machine learning algorithms to implement learning, have been particularly instrumental in targeted marketing strategies, enhance customer this regard, as they do not require predefined satisfaction, and