Table 1: Metrics.

Metric

Description

Formula

TPR

(sensitivity or recall)

True Positive Rate

·         probability of a positive test, given a patient ill;

·         measure of prediction correctness;

·         should be high.

 

PPV

(or precision)

Positive Predictive Value

·         the proportion of the true positives against all the positive results;

·         capacity of predicting the positives;

·         used instead of FPR;

·         should be high.

F-measure

 

F-measure

·         harmonic weighted mean between PPV and TPR;

·         close to 1 = better prediction.

MCC

 

 

 

Matthew’s Correlation Coefficient

·         equal to:

- -1: false assignments (TN = TP = 0)

-  0: prediction not better than random

-  1: all true assignments (FP = FN = 0)

Set to 0 when denominator = 0

·         ~ geometric mean between TPR and PPV;

·         should be high.

 

TP = True Positive; TN = True Negative; FN = False Negativ; FP = False Positive