Unverified50% confidenceOpinionExact time
When the proportion of diseased samples is low, recall and F1 score reflect the model's actual detection capability better than accuracy.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFine-tuned DistilBERT achieves F1=0.9993 on in-distribution testing, but loses 12.9 mean average precision points under topic shift, a greater degradation than the linear model's 5.2 points.73% similarUnverifiedEven the best-performing model achieves an evidence ID F1 of only around 0.36, reflecting that current large models still have significant room for improvement on cross-disciplinary reasoning tasks requiring precise evidence localization.73% similarUnverifiedIn imbalanced classification problems, a model can achieve very high accuracy even by predicting all samples as the majority class, so it is necessary to shift to more discriminative metrics such as precision-recall curves, F1 score, Matthews correlation coefficient, or AUC-ROC71% similarUnverifiedF1 levels between 70 and 80 indicate that current model performance is more suitable as a research reference than as a directly deployable decision-assistance tool.71% similarUnverified在疟疾筛查等高风险医疗场景中，假阴性的后果远比假阳性危害大，因此模型设计倾向于以较低决策阈值换取高召回率，而非追求F1或AUC-ROC的最优70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1028197API
curl https://kongchang.com/api/v1/knowledge/claims/1028197MCP
get_claim(id=1028197)