彼尔·约翰逊 , 邓曼瑶 , 高瑞泽 , 韩立岩 , 李振 , 刘瑾 , 龙飞 , 史冬波 , 汤珂 , 严兴 , 张琦

陈秀萍

2026年09月18日 15:00 至 16:00

A3-2-303

Zoom 242 742 6089 (BIMSA)

This paper proposes XPER, a Shapley value-based framework for measuring the driving forces of predictive performance in credit scoring models. Unlike existing explainable methods that mainly focus on individual predictions, XPER decomposes overall model performance metrics (e.g., AUC and R2) into feature-level contributions. The method is model-agnostic and can be applied to different predictive models and performance measures. Using car loan data, the authors show that a small number of features account for a large share of predictive performance, while the key drivers of overall predictive ability may differ from those affecting individual predictions.

Xiuping Chen is an Intern at BIMSA. Her research interests include digital economy, data assets and commercial bank.