FairLoop：面向以人为中心公平性的预测业务流程监控软件支持
机器学习
2025-08-28 v1
摘要
敏感属性如性别或年龄在机器学习任务中如预测业务流程监控时，若不考虑情境会导致不公平的预测结果。我们提出FairLoop1，一个用于神经网络预测模型的人类引导偏差缓解工具。FairLoop从神经网络蒸馏出决策树，使用户能够检查和修改不公平的决策逻辑，然后用于针对更公平的预测进行微调。与其他公平性方法相比，FairLoop通过人类参与实现情境感知的偏差消除，针对性地处理敏感属性的影响而非统一排除。
引用
@article{arxiv.2508.20021,
title = {FairLoop: Software Support for Human-Centric Fairness in Predictive Business Process Monitoring},
author = {Felix Möhrlein and Martin Käppel and Julian Neuberger and Sven Weinzierl and Lars Ackermann and Martin Matzner and Stefan Jablonski},
journal= {arXiv preprint arXiv:2508.20021},
year = {2025}
}
备注
Proceedings of the Best BPM Dissertation Award, Doctoral Consortium, and Demonstrations & Resources Forum co-located with 23rd International Conference on Business Process Management (BPM 2025), Seville, Spain, August 31st to September 5th, 2025