1.ClassLabels:类型标识.第一个label作为pos,第二次label作为neg.

2.GroundTruth:各次实验的观察值,也就是真实值.

3.ValidationCounter: 测试次数

4.SampleDistribution:每个样本作为测试集样本的次数.如果是k-fold-validation则会有k次.

5.ErrorDistribution:在测试时每个样本被误判的次数

以上2个属性在k-fold-valiation中可以找出误判次数多的样本.

6.SampleDistributionByClass:在测试集中各类样本数量

7.ErrorDistributionByClass:在被误判的样本集中各类样本的数量

8.CountingMatrix:前2行表示TP,FP;TN,FN;最后一行是inconclusive results.

9.CorrectRate: (TP+FN)/(P+N)

10.ErrorRate:1-CorrectRate

11.Sensitivity: TP/(TP+FP)=recall=FDR(Failure detective rate)

12.Specificity: FN/(TN+FN)=1-FAR(false alarm rate)

13.PositivePredictiveValue:TP/(TP+TN)=precision

14.NegativePredictiveValue:FN/(FP+FN)

15.Prevalence:TP/(TP+FP+TN+FN)

16.DiagnosticTable:与CountingMatrix相同

要求recall,precision,FAR,FDR可以直接在取.

原文：http://www.cnblogs.com/york-hust/p/3962270.html