y

[

,

;

ez-Orallo, et al., 2013].

(a) (b)

a) The total prediction accuracy curve of threshold variation for the seeds data.

t curve of threshold variation for the seeds data.

OCR package also provides a function for extracting the optimal

through the cost analysis. The optimal threshold for the

m cost can also be extracted. Figure 3.17(b) shows the cost curves

d with the threshold variation for two models. The optimal

for minimising the cost can be extracted and they were 0.757 for

-validation model and 0.563 for the Jackknife model. All were

from 0.5. This indicated that optimal discrimination threshold

ecessary at 0.5 all the time. In other words, the optimal threshold

y from 0.5 depending on a specific data set and a different

ation test approach for a data set.

are also other useful packages for the ROC analysis. The pROC

can be used for the ROC analysis with a confidence interval

t al., 2011]. The package employs the bootstrap approach [Efron ,

eiman, 1996] to estimate the confidence bands for an ROC curve

lso be used for the partial ROC analysis. The pROC package does

he false alarm rate as the horizontal axis. Instead, it uses the

y as the horizontal axis. This is why its horizontal axis ranges

to zero. Figure 3.18 shows the ROC curves for the seeds data

OC.