ontal axis of an ROC curve and measure stands for the vertical

n ROC curve,

curve=performance(my.roc,measure='tpr',

x.measure='fpr')

d on my.curve, the plot function is called to generate an ROC

o acquire the AUC, the performance function is called again,

uc must be specified for the measure parameter,

auc.obj=performance(my.roc,measure='auc')

AUC value can thus be extracted from this AUC object using the

g R code, where my.auc is the calculated AUC value and the use

y is because my.auc.obj is an S4 object,

my.auc=unlist(my.auc.obj@y.values)

neralisation

rtant issue associated with a supervised machine learning model

neralisation power. The generalisation power of a supervised

learning model is an estimation of how a constructed supervised

learning model can work well for new data. Whether a model

ell for new data is much more important than whether a model

ell for data which is used to construct the model. A process of

e generalisation power of a supervised machine learning model

he generalisation test, which has to be implemented on new data.

ning of new data is that the data has not been used for model

ion.

lassification models are supervised machine learning models,

rtainly require a generalisation test in place [Devroye, et al., 1996;

1996; Ripley, 1996]. The subject of the generalisation test of a

d machine earning model has been well-exercised for analysing

l/medical data as well. For instance, classification models have

structed for gastric and colonic epithelial tumour [Iizuka, et al.,