An explainable artificial intelligence approach for financial distress prediction
作者：
Highlights：
• Propose a new explainable artificial intelligence approach for financial distress prediction.
• Take financial distress prediction as an applied setting environment, analyze who the external stakeholders are and what their interpretative desires are, and then establish an explainable framework.
• Apply shapley additive explanations, partial dependence plots and counterfactual explanations to generate local explanations and global explanations to improve the transparency and utilization of “black box” model.
• Implement a whole process ensemble model from feature selection to predictor construction to maximize the effectiveness of ensemble.
• Design a two-stage ensemble model integrated with a filter and wrapper technique for feature selection.
摘要
•Propose a new explainable artificial intelligence approach for financial distress prediction.•Take financial distress prediction as an applied setting environment, analyze who the external stakeholders are and what their interpretative desires are, and then establish an explainable framework.•Apply shapley additive explanations, partial dependence plots and counterfactual explanations to generate local explanations and global explanations to improve the transparency and utilization of “black box” model.•Implement a whole process ensemble model from feature selection to predictor construction to maximize the effectiveness of ensemble.•Design a two-stage ensemble model integrated with a filter and wrapper technique for feature selection.
论文关键词：Financial distress prediction,Explainable artificial intelligence,Ensemble method,Feature selection,SHapley Additive exPlanations
论文评审过程：Received 22 December 2021, Revised 24 April 2022, Accepted 28 May 2022, Available online 7 June 2022, Version of Record 7 June 2022.
论文官网地址：https://doi.org/10.1016/j.ipm.2022.102988
