Forecasting loan, deferred rate and customer segmentation in banking industry: A computational intelligence approach
Accurate loan default prediction and customer segmentation are critical challenges in the banking industry. This study proposes a novel hybrid model integrating Multi-Layer Perceptron (MLP) neural networks with Self-Adaptive Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) frameworks. GA...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-09-01
|
| Series: | Array |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2590005625000876 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate loan default prediction and customer segmentation are critical challenges in the banking industry. This study proposes a novel hybrid model integrating Multi-Layer Perceptron (MLP) neural networks with Self-Adaptive Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) frameworks. GA handles feature selection, while PSO optimizes MLP hyperparameters (e.g., learning rate, neurons, activation functions). The model dynamically enhances classification accuracy and resilience, particularly for imbalanced datasets in loan default prediction. Using real-world data from Sina Bank, the system outperforms Logistic Regression, Decision Trees, and Random Forests. The GA-PSO optimization process, which integrates both PSO and GA to optimize the MLP model's parameters, plays a crucial role in enhancing the accuracy and scalability of the system. Specifically, the GA-PSO-MLP model achieves a 15 % higher classification accuracy than Logistic Regression, a 12 % improvement over Decision Trees, and an 8 % gain over Random Forests. Additionally, false positive rates are reduced by 20 %, and mean squared error (MSE) is lowered by 18 %. The F1-score of the proposed model is 92.3 %, compared to 79.8 % (Logistic Regression), 81.5 % (Decision Trees), and 85.2 % (Random Forests), further highlighting its advantage in handling imbalanced datasets. Extensive numerical validation and sensitivity analysis further highlight the model's effectiveness in delivering actionable insights that enhance customer management strategies and mitigate financial risks. This research makes a substantial contribution to the application of machine learning in banking, facilitating more accurate data-driven decision-making and more robust risk management practices. |
|---|---|
| ISSN: | 2590-0056 |