Q-rung picture fuzzy sets theory-based ensemble learning method for credit risk assessment
Credit risk assessment is conducted to assess the credit repayment capacity of a customer and determine the associated risk level. Traditional methods often face limitations in the effective modeling of uncertainties and the management of large datasets, highlighting the need to improve the risk management processes of banks. In addition, evaluating credit risk is essential to guide decision-making mechanisms within financial organizations; however, traditional approaches often struggle with the complexity and uncertainty involved. q-rung picture fuzzy sets (q-RPFSs), a theory well known for its ability to handle uncertainty, offer significant potential to address these challenges. In this study, we propose a novel framework that integrates q-RPFS-based feature selection with a stacking ensemble learning method to enhance predictive performance. The framework was tested on the German credit dataset. The findings demonstrated the effectiveness of the method, achieving an accuracy of 0.768 and an AUC of 0.792 on the German dataset, placing it among the top-performing approaches in the literature. Additionally, comparative experiments were conducted using several widely adopted fuzzy-MCDM methods. These included fuzzy Analytic Hierarchy Process (fuzzy-AHP), fuzzy Best Worst Method (fuzzy-BWM), and fuzzy Criteria Importance Through Intercriteria Correlation (fuzzy-CRITIC). The results indicated that the proposed q-RPFSs-based approach consistently outperforms these alternative techniques in terms of predictive performance. As a result, by integrating q-RPFSs with machine learning, this approach not only introduces innovative solutions to the financial sector but also provides a robust framework for handling uncertainty in real-world decision-making, thereby advancing the state of the art in intelligent systems.
- Journal
- KNOWLEDGE-BASED SYSTEMS
- Year
- 2026
- Volume
- 341
- ISSN
- 0950-7051
- eISSN
- 1872-7409
- DOI
- 10.1016/j.knosys.2026.115581
- WoS UID
- WOS:001728423300001