allows the partial ROC analysis. Suppose the false alarm rate was
o 0.05. This means that any threshold which may result in more
false alarm rate was not considered. Figure 3.16(b) shows this
OC analysis. The partial AUC values were also smaller than 0.05.
re 0.0062 and 0.005 for using the cross-validation approach and
nife test approach.
The left panel is the confusion matrix of the cross-validation LDA model for the
The right panel is the confusion matrix of the Jackknife test LDA model for the
Cross-validation
Jackknife test
A
B
%
A
B
%
A
62
9
87.3
63
8
88.7
B
4
61
93.8
4
61
93.8
%
93.9
87.1
89.3
94.0
88.4
90.0
(a) (b)
ROC curves of the cross-validation and Jackknife LDA models for the seeds data
R. (a) Full ROC curves. (b) Partial ROC curves.
ROCR package can provide the analysis of the total prediction
associated with the threshold variation. It measures how the total
n accuracy of a classifier varies when the threshold is varied.
o the ROC analysis, varying the threshold can generate a total
n accuracy curve. Figure 3.17(a) shows a total prediction
curve for the threshold variation for the seeds data. Using
