sideChannelAttack (version 1.0-6)

verify.cv: verify.cv

Description

The verify.cv function allows to estimate the quality of a model (an attack) with a k-cross-validation approach.

Usage

verify.cv(model, filter, X,Y, nbreVarX, k, param.model=list(), param.fs=list(), ...)

Arguments

model

A model such as randomForest, gaussian, svm, etc.

filter

A feature selection such as filter.PCA, filter.mRMR, etc.

X

A matrix where each row is a physical measure.

Y

A vector where the $i_th$ element of the vector $y$ is the key for the $i_th$ physical measure in the matrix $x$.

nbreVarX

The number of variables which represents each physical measure after the reduction by the feature selection.

k

The number of traces in the validation set.

param.model

A list of parameters to insert into the model.

param.fs

A list of parameters to insert into the feature selection algorithm.

...

Currently ignored.

Value

The verify.cv function returns an object which can be used with the plot function to plot the quality of the model.

TP

number of true positive

TN

number of true negative

FN

number of false negative

FP

number of false positive

dim

the number of variables which represents each physical measure after the reduction by the feature selection.

Details

The verify.cv function estimates the quality of the attack with a k-cross-validation approach. It is executed in $N/k$ rounds. Each round uses $N-k$ traces to learn a model and the remaining trace to assess the generalization accuracy of the model. This is repeated until every set of k traces has been used for testing purposes. The best model is the one that maximizes the value returned by k-cross-validation.

Examples

#data collection
data(powerC)
traces = powerC[,-301]
traces = traces[,1:10]
key = powerC[,301]
newIndice = c(sample(1:128,15),sample(129:256,15))
traces = traces[newIndice,]
key = key[newIndice]+1

#model checking
attack=verify.cv(model=dpa1,filter=filter.PCA,X=traces,Y=key,nbreVarX=c(2:4),k=2)
plot(attack)