arXiv：cs.LG· Wenhao Chi, \c{S}. \.Ilker Birbil·· 5 小时前AI 评分33
面向决策净收益最大化的可解释风险评分系统
Learning an Interpretable Risk Scoring System for Maximizing Decision Net Benefit
AI 导读
研究者提出一种直接优化多决策阈值下净收益的风险评分系统，将其建模为稀疏整数线性规划问题，从而构建整数系数的透明评分体系。该方法还推导出净收益曲线下面积与 ROC 泛函在固定阈值网格上的界，并证明后处理可在不降低该网格净收益曲线下面积的前提下实现训练数据上的适度校准。在多个公开数据集及一个大规模信用风险数据集上的实验显示，该方法在保持有竞争力的区分度与校准性能的同时实现了高净收益。
正文
Abstract:Risk scoring systems are widely used in high-stakes domains to assist decision-making. However, existing approaches often focus on optimizing predictive accuracy or likelihood-based criteria, which may not align with the main goal of maximizing utility. In this paper, we propose a novel risk scoring system that directly optimizes net benefit over a range of decision thresholds. The model is formulated as a sparse integer linear programming problem which enables the construction of a transparent scoring system with integer coefficients, and hence, facilitates interpretation and practical application. We also establish fundamental relationships among net benefit, discrimination, and calibration. Specifically, we derive bounds relating the area under the net benefit curve to a ROC functional, both evaluated on a fixed threshold grid, and show that post-processing can achieve moderate calibration on the training data without decreasing the area under the net benefit curve on that grid. We evaluated our method on multiple public datasets as well as on a large-scale credit risk dataset. This computational study demonstrated that our interpretable method can effectively achieve high net benefit while maintaining competitive discrimination and calibration performance.
| Comments: | 53 pages, 9 figures, 18 tables, and 6 algorithms |
| Subjects: | Machine Learning (cs.LG); Optimization and Control (math.OC) |
| Cite as: | arXiv:2604.04241 [cs.LG] |
| (or arXiv:2604.04241v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2604.04241 arXiv-issued DOI via DataCite |
Submission history
From: Wenhao Chi [view email]
[v1]
Sun, 5 Apr 2026 19:53:29 UTC (141 KB)
[v2]
Tue, 14 Jul 2026 13:07:47 UTC (182 KB)
[v3]
Fri, 2 Oct 2026 15:19:33 UTC (170 KB)
来源：arXiv：cs.LG · arxiv.org