Example dependent cost sensitive learning based selective deep ensemble model for customer credit scoring.

Area of Science:

• Machine Learning
• Data Science
• Financial Analytics

Background:

• Class-imbalanced data is a common challenge in credit scoring.
• Traditional cost-sensitive methods often fail to account for varying sample costs and lack real-world applicability.
• Limited interpretability of existing models hinders practical adoption in financial decision-making.

Purpose of the Study:

• To propose a novel example-dependent cost-sensitive learning based selective deep ensemble (ECS-SDE) model for enhanced customer credit scoring.
• To address limitations of traditional methods by integrating varying costs and improving model interpretability.
• To develop a credit scoring solution that aligns better with business needs and provides transparent decision-making.

Main Methods:

• Developed an ECS-SDE model integrating example-dependent cost-sensitive learning with TabNet (attentive interpretable tabular learning) and GMDH (group method of data handling) deep neural networks.
• Utilized TabNet as the base classifier, optimizing it for imbalanced data using an example-dependent cost loss function.
• Designed a GMDH with an example-dependent cost-sensitive symmetric criterion for selective deep integration of base classifiers, reducing redundancy and enhancing performance.

Main Results:

• The ECS-SDE model demonstrated superior overall performance compared to six cost-sensitive and five advanced deep ensemble models in credit scoring.
• Achieved significant advantages in BS+, Save, and AUC metrics across four datasets.
• The model provided strong interpretability, with detailed analysis highlighting key feature roles in credit scoring.

Conclusions:

• The proposed ECS-SDE model effectively addresses class imbalance and varying costs in credit scoring data.
• ECS-SDE offers improved classification performance and enhanced interpretability over existing methods.
• This approach provides a more robust and transparent solution for practical credit scoring applications.