机器学习-分类度量(classification metric)常用评价指标
分类:
文章
•
2022-10-03 16:39:11
- 评判指标存在的目的
- 应用场景
- 混淆矩阵
- accuracy =(TP+TN)/(TP+TN+FP+FN)
- precision =TP/(TP+FP)
- recall 召回率=真阳性率(True Positive Rate,TPR)=灵敏度(Sensitivity)=(TP/TP+FN)
- P-R曲线=precision recall curve
- 真阴性率(True Negative Rate,TNR),特异度(Specificity)=TN/(TN+FP)
- F1-Score=2precisionrecall/(precision+recall)
- 假阴性率(False Negatice Rate,FNR)=1 - 灵敏度(Sensitivity)=FN/(TP+FN)
- 假阳性率(False Positice Rate,FPR)= 1 - 特异度(Specificity)=FP/(FP+TN)
- ROC曲线=接收者操作特征曲线(receiver operating characteristic curve)
- 曲线下面积(Area Under Curve)
- 多分类问题