A CWGAN-GP-based multi-task learning model for consumer credit scoring

作者：

Highlights：

• We propose a CWGAN-GP-Based Multi-task Learning Model for credit scoring.

• Adjust the distribution between good and bad data by augmenting synthetic bad data.

• A MTL framework on both accepted and rejected and good and bad data is proposed.

• The proposed model is evaluated on different real-world loan datasets.

• The empirical results indicate the proposed model achieves better performance.

摘要

•We propose a CWGAN-GP-Based Multi-task Learning Model for credit scoring.•Adjust the distribution between good and bad data by augmenting synthetic bad data.•A MTL framework on both accepted and rejected and good and bad data is proposed.•The proposed model is evaluated on different real-world loan datasets.•The empirical results indicate the proposed model achieves better performance.

论文关键词：Credit scoring,Deep learning,Generative adversarial networks,Imbalanced data,Multi-task learning

论文评审过程：Received 22 June 2021, Revised 2 May 2022, Accepted 27 May 2022, Available online 6 June 2022, Version of Record 16 June 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117650