Evaluation metrics for classification

Accuracy/Error rate

ACC = (TP+TN)/(P+N)

 

ERR = (FP+FN)/(P+N) = 1-ACC

 

Confusion matrix

Precision/Recall/F1

Precision = TP/(TP+FP)-- positive predictive value

 

Recall= TP/(TP+FN)  -- true positive rate

 

F1=1/(1/precision+1/recall)

 

ROC

True positive rate (TPR): the ratio of positive instances that are correctly classified as positive

TPR = TP/(TP+FN) = recall

 

True negative rate (TNR): the ratio of negative instances that are correctly classified as negative

TNR = TN/(TN+FP) = specify

 

False positive rate (FPR): the ratio of negative instances that are incorrectly classified as positive. 

FPR = FN/(TN+FP) = 1-specify

 

ROC: TPR vs FPR

 

Matthews correlation coefficient

 

 

Logarithm loss/cross entropy