Unverified50% confidenceFactExact time
当正负样本比例极端时，PR曲线下的面积（AUPRC）往往比AUC更能揭示真实性能
1
Sources
50%
Confidence
Long-term
Relevance
7/12/2026
First Seen
Sources
从零训练机器学习模型：新手完整入门指南
redditr/learnmachinelearning7/11/2026
Related Claims
Unverified在极度不平衡数据集中（阳性率低于5%），PR曲线下面积（AP）往往比AUC更能反映模型对少数类的检测能力83% similarUnverifiedAUC-ROC的核心优势在于对类别不平衡不敏感，能反映模型在不同分类阈值下的综合区分能力，AUC值范围在0.5到1.0之间73% similarUnverifiedAUC的取值范围在0到1之间，数值越大代表模型分类能力越强72% similarUnverifiediPSAE评分对Binder二分类预测的AUC仅为0.65，远低于可靠分类器水平69% similarUnverifiedAUC（Area Under Curve）是衡量分类模型好坏最常用的指标之一68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/490198API
curl https://kongchang.com/api/v1/knowledge/claims/490198MCP
get_claim(id=490198)