Skip to content
arXiv · Statistics Machine Learning· Sunny Yang, Weiyan Zhao·· 4 hr agoAI score24

竞争风险下的失效原因预测集:何时需要删失权重?

When does conformal calibration need censoring weights? Cause-of-failure prediction sets under competing risks

AI brief

竞争风险标签的 split conformal 预测集在校准时可能因右删失而缺失标签。仿真显示,22% 受试者在目标时点无事件时,complete-case 校准在名义 0.900 覆盖率下仅达 0.8723;正确设定的删失模型权重可让覆盖率接近名义水平。论文还给出含删失模型误差惩罚的有限样本覆盖率下界。

Source: arXiv · Statistics Machine Learning · arxiv.org