Document Type : Original Research Paper
Authors
1 Department of Management, ToH.C., Islamic Azad University, Torbat Heydarieh, Iran.
2 Department of Economics, Management and Accounting, Payame Noor University, Tehran, Iran.
Abstract
Background and Objectives: The rapid growth of digital banking, electronic payment systems, and online financial services has generated vast amounts of transactional and customer-related data. These developments have created new opportunities for data-driven decision-making in the banking sector, particularly in demand forecasting, service planning, and branch network optimization. Because banking demand is influenced by spatial, demographic, and socioeconomic factors, integrating geospatial analysis with artificial intelligence techniques can provide valuable insights for strategic banking decisions. Despite the increasing availability of banking data, the practical use of spatial data mining and machine learning for demand analysis and branch planning remains limited. Therefore, this study aims to develop and evaluate an integrated framework that combines spatial analysis, machine learning, deep learning, and explainable artificial intelligence to forecast banking service demand and support spatial decision-making for branch network optimization.
Methods: The study was conducted using customer and branch data obtained from Bank Sepah in Markazi Province, Iran, covering the period from 2021 to 2023. Markazi Province was selected as the study area because of its diverse urban, industrial, rural, and service-oriented characteristics, which provide a suitable environment for investigating heterogeneous patterns of banking demand. The proposed framework consists of four main stages. First, transactional, spatial, and socioeconomic datasets were integrated and processed through data cleaning, preprocessing, and feature engineering procedures. Second, spatial demand patterns were investigated using exploratory spatial data analysis, including Moran’s I index, Local Indicators of Spatial Association, and density-based spatial clustering methods. Third, predictive models were developed using Random Forest, Extreme Gradient Boosting, Long Short-Term Memory networks, and Spatio-Temporal Graph Neural Networks. Finally, model outputs were interpreted using feature importance analysis and Shapley Additive Explanations to identify the key drivers of banking demand and support decision-making processes. Model performance was evaluated using standard statistical measures, including the coefficient of determination, precision-recall analysis, calibration assessment, and classification performance metrics.
Findings: The results revealed significant spatial heterogeneity in banking service demand across the study area. Spatial analysis produced a Moran’s I value of 0.71, indicating strong positive spatial autocorrelation and confirming the existence of clustered demand patterns. The identified hotspots were primarily concentrated in economically active and densely populated counties. Comparative evaluation of predictive models demonstrated that tree-based machine learning algorithms outperformed deep learning approaches for the available dataset. Random Forest achieved the highest predictive performance with an R² value of 0.95, followed closely by Extreme Gradient Boosting with an R² value of 0.94 for forecasting monthly customer visits. In contrast, the Spatio-Temporal Graph Neural Network model achieved a lower performance level (R² ≈ 0.51). Explainability analysis showed that spatial accessibility, customer socioeconomic characteristics, and historical transaction behavior were among the most influential factors affecting demand patterns. The results further demonstrated that the integration of spatial features substantially improved predictive accuracy and enhanced the interpretability of the generated models.
Conclusion: The findings indicate that combining spatial analytics, machine learning, and explainable artificial intelligence provides an effective framework for understanding and forecasting banking service demand. Beyond model comparison, the proposed framework supports practical decision-making related to branch network development, service allocation, and digital banking expansion. The identified spatial demand patterns can assist banking managers in prioritizing investment locations, improving service coverage, and optimizing operational resources. However, the limited temporal depth of the available dataset restricted the effectiveness of advanced spatio-temporal deep learning models. Future studies may benefit from longer time-series datasets and real-time transactional information to further enhance predictive performance and spatio-temporal modeling capabilities. Overall, the proposed framework offers a practical and scalable approach for developing intelligent decision-support systems in data-driven banking environments.
Keywords
- Machine Learning
- Smart Banking
- Demand Forecasting
- Model Explainability
- Spatio-Temporal Graph Neural Networks
Main Subjects
COPYRIGHTS
© 2026 The Author(s). This is an open-access article distributed under the terms and conditions of the Creative Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)
)