Optimizing Multiclass Banking Customer Segmentation: A Dual Approach using Transformer Model and Advanced Optimizers

Abstract

Financial institutions and banking sectors, enhance customer satisfaction, improve customer retention rates, and optimize revenue of the business through customer segmentation. Customer segmentation is a key factor to consider in the business; today’s business world is struggling to handle their customers in an effective way. As the business can create a competitive edge by focusing more on customer than the product. While identifying the loyal customers, they can offer more product offerings, tailored service, and improve the business revenue. Consequently, for the effective customer segmentation, the artificial intelligence, machine learning and deep learning techniques played a major role in the present day. Through deep learning techniques the model learned the customer behavior in many aspects and categorized, however, this architecture proved reliable models for customer segmentation. To handle the complexity and diversity of the customer information, traditional machine learning approaches often struggle. Hence, the research explores the deep learning models – Artificial Neural Networks (ANN), Gated Recurrent Units (GRU), and Transformer Networks (TN), for multiclass customer classification. ANN is a powerful technique to learn effectively the data with complex non-linear pattern, while GRU are effective in handling the sequential data, and TN are excellent for self-attention mechanisms. Based on the metrics such as accuracy, precision, recall, and f1-score, all models performs well. Among these, transformer model excelled others in capturing complex data patterns, and providing superior segmentation. The study explores the efficacy of advanced deep learning architectures for better customer prediction and segmentation; in that way predict the profitable customer and an upturn in the economy.