ve measurements are defined as below,

ecificity

TN

TN ൅FP

nsitivity

TP

TP ൅FN

al accuracy

TN ൅TP

TN ൅FP ൅FN ൅TP

gative prediction power

TN

TN ൅FN

sitive prediction power

TP

TP ൅FP

(3.28)

ceiver operating characteristic analysis

ion matrix is a fix-point evaluation approach. It does not measure

st a classifier is. In most medical applications, it is often required

threshold so as to minimise the cost involved with the life loss.

sure that missing a cancer patient diagnosis has a much greater

a falsely diagnosed cancer patient.

e 3.10 shows such an example of the impact of a threshold on the

tion performance. There are two densities for two classes from a

ed classifier in this figure. The left density curve is for one class

ght density curve is for the other class. Two densities have been

ed meaning the misclassification will more or less happen no

hat a threshold is used to separate two classes. For two thresholds

the dots on the horizontal axes, the misclassification rates shown

aded areas in two panels are different. In Figure 3.10(a), the left

a greater misclassification rate than the right class. The grey

rea (for the left class) is much larger than the dark shaded area

ight class). But in Figure 3.10(b), the right class has a greater

fication rate than the left class. The dark shaded area is much

an the grey shaded area. This is because the threshold has been

owards the right direction. This shows that classification