[R] bagging SVM Ensemble

[Iut Tri Utami]

Dear Sir,

I got a problem with my program. I would like to classify my data using
bagging support vector machine ensemble. I split my data into training data
and test data. For a given data sets TR(X), K replicated training data sets
are first randomly generated by bootstrapping technique with replacement.
Next, Support Vector Mechine (SVM) is applied for each bootstrap data sets.
Finally, the estimation of bagging SVM is obtained by aggregating
independently trained SVM’s in an appropriate aggregation technique.I've
tried the program, can you correct them? I have the results of the program in
accordance with the appropriate I expected because it turns out I do not get
the missclassification Rate and aggregation. Can you help me??

thanks you so much for your attention

best wishes,

IUT TRI UTAMI
Bogor Agricultural Institute, Indonesia
-------------- next part --------------
#SINGLE SVM
library(colorspace)
library(rpart)
library(e1071)
library(MASS)
dataset <- read.csv("E:/thesis/SVM/hepatitis data csv.csv",header=T,sep=";")
attach(dataset)
index <- 1:nrow(dataset)
testindex <- sample(index, trunc(length(index)*30/100))
testset <- dataset[testindex,]
trainset <- dataset[-testindex,]
trainindex <- sample(index, trunc(length(index)*70/100))
tuned <- tune.svm(class~., data = trainset, gamma = 10^(-6:-1), cost = 10^(-1:1))
cc <- as.numeric(tuned$best.parameters[2])
gg <- as.numeric(tuned$best.parameters[1])
modelolin <- svm(class ~ ., trainset, type = "C-classification", cost = cc, gamma = gg, kernel = "linear")
modelopoly <- svm(class ~ ., trainset, type = "C-classification", cost = cc, gamma = gg, kernel = "polynomial")
modelorad <- svm(class ~ ., trainset, type = "C-classification", cost = cc,gamma = gg, kernel = "radial")
modelotan <- svm(class ~ ., trainset, type = "C-classification", cost = cc,gamma = gg, kernel = "sigmoid")
prediction1<- predict(modelolin,newdata=testset, decision.values=F)
prediction2<- predict(modelopoly,newdata=testset, decision.values=F)
prediction3<- predict(modelorad,newdata=testset, decision.values=F)
prediction4<- predict(modelotan,newdata=testset, decision.values=F)
tab1 <- table(pred = prediction1,testset$class)
tab2 <- table(pred = prediction2,testset$class)
tab3 <- table(pred = prediction3,testset$class)
tab4 <- table(pred = prediction4,testset$class)
ecr1<-(tab1[2,1]+tab1[1,2])/sum(tab1)
ecr2<-(tab2[2,1]+tab2[1,2])/sum(tab2)
ecr3<-(tab3[2,1]+tab3[1,2])/sum(tab3)
ecr4<-(tab4[2,1]+tab4[1,2])/sum(tab4)

classAgreement(tab)


#Bagging EnSVM
library(colorspace)
library(rpart)
library(e1071)
library(MASS)
dataset <- read.csv("E:/thesis/SVM/hepatitis data csv.csv",header=T,sep=";")
 attach(dataset)
 m<-10
 k<-10
index <- 1:nrow(dataset)
 testindex <- sample(index, trunc(length(index)*30/100))
 testset <- dataset[testindex,]
 trainset <- dataset[-testindex,]
trainindex<- sample(index, trunc(length(index)*70/100))
 es<-numeric(m)
 eb<-matrix(nrow=m,ncol=k)
 eslin<-numeric(m)
 espoly<-numeric(m)
 esrad<-numeric(m)
 estan<-numeric(m)
eblin<-matrix(nrow=m,ncol=k)
ebpoly<-matrix(nrow=m,ncol=k)
ebrad<-matrix(nrow=m,ncol=k)
ebtan<-matrix(nrow=m,ncol=k)
set.seed(123)
for (i in 1:m){
index <- 1:nrow(dataset)
 testindex <- sample(index, trunc(length(index)*30/100))
 testset <- dataset[testindex,]
 trainset <- dataset[-testindex,]
 tuned <- tune.svm(class~., data = trainset, gamma = 10^(-6:-1), cost = 10^(-1:1))
cc <- as.numeric(tuned$best.parameters[2])
gg <- as.numeric(tuned$best.parameters[1])
modellin <- svm(class ~ ., trainset, type = "C-classification", cost = cc, gamma = gg, kernel = "linear")
modelpoly <- svm(class ~ ., trainset, type = "C-classification", cost = cc, gamma = gg, kernel = "polynomial")
modelrad <- svm(class ~ ., trainset, type = "C-classification", cost = cc,gamma = gg, kernel = "radial")
modeltan <- svm(class ~ ., trainset, type = "C-classification", cost = cc,gamma = gg, kernel = "sigmoid")
prediction1<- predict(modellin,newdata=testset, decision.values=F)
prediction2<- predict(modelpoly,newdata=testset, decision.values=F)
prediction3<- predict(modelrad,newdata=testset, decision.values=F)
prediction4<- predict(modeltan,newdata=testset, decision.values=F)
no1<-sum(prediction1==0)
yes1<-sum(prediction1==1)
vote1<-c()
if(yes1>no1) vote1<-1 else vote1<-0
no2<-sum(prediction2==0)
yes2<-sum(prediction2==1)
vote2<-c()
if(yes2>no2) vote2<-1 else vote2<-0
no3<-sum(prediction3==0)
yes3<-sum(prediction3==1)
vote3<-c()
if(yes3>no3) vote3<-1 else vote3<-0
no4<-sum(prediction4==0)
yes4<-sum(prediction4==1)
vote4<-c()
if(yes4>no4) vote4<-1 else vote4<-0
tab1 <- table(pred = prediction1,testset$class)
tab2 <- table(pred = prediction2,testset$class)
tab3 <- table(pred = prediction3,testset$class)
tab4 <- table(pred = prediction4,testset$class)
ecr1<-(tab1[2,1]+tab1[1,2])/sum(tab1)
ecr2<-(tab2[2,1]+tab2[1,2])/sum(tab2)
ecr3<-(tab3[2,1]+tab3[1,2])/sum(tab3)
ecr4<-(tab4[2,1]+tab4[1,2])/sum(tab4)
eslin[i]<-ecr1
 espoly[i]<-ecr2
 esrad[i]<-ecr3
 estan[i]<-ecr4
set.seed(123)
	for (j in 1:k) {
	boot<-sample(trainindex,replace=TRUE)
	sboot<-trainset[boot,]
	tuned <- tune.svm(class~., data = sboot, gamma = 10^(-6:-1), cost = 10^(-1:1))
cc <- as.numeric(tuned$best.parameters[2])
gg <- as.numeric(tuned$best.parameters[1])
modellin1 <- svm(class ~ ., sboot, type = "C-classification", cost = cc, gamma = gg, kernel = "linear")
modelpoly2 <- svm(class ~ ., sboot, type = "C-classification", cost = cc, gamma = gg, kernel = "polynomial")
modelrad3 <- svm(class ~ ., sboot, type = "C-classification", cost = cc,gamma = gg, kernel = "radial")
modeltan4 <- svm(class ~ ., sboot, type = "C-classification", cost = cc,gamma = gg, kernel = "sigmoid")
predictions1<- predict(modellin1,newdata=testset, decision.values=F)
predictions2<- predict(modelpoly2,newdata=testset, decision.values=F)
predictions3<- predict(modelrad3,newdata=testset, decision.values=F)
predictions4<- predict(modeltan4,newdata=testset, decision.values=F)
noboot1<-sum(predictions1==0)
yesboot1<-sum(predictions1==1)
voteboot1<-c()
if(yesboot1>noboot1) voteboot1<-1 else voteboot1<-0
noboot2<-sum(predictions2==0)
yesboot2<-sum(predictions2==1)
voteboot2<-c()
if(yesboot2>noboot2) voteboot2<-1 else voteboot2<-0
noboot3<-sum(predictions3==0)
yesboot3<-sum(predictions3==1)
voteboot3<-c()
if(yesboot3>noboot3) voteboot3<-1 else voteboot3<-0
noboot4<-sum(predictions4==0)
yesboot4<-sum(predictions4==1)
voteboot4<-c()
if(yesboot4>noboot4) voteboot4<-1 else voteboot4<-0
tabs1 <- table(pred = predictions1,testset$class)
tabs2 <- table(pred = predictions2,testset$class)
tabs3 <- table(pred = predictions3,testset$class)
tabs4 <- table(pred = predictions4,testset$class)
ecrbootmodellin<-(tabs1[2,1]+tabs1[1,2])/sum(tabs1)
ecrbootmodelpoly<-(tabs2[2,1]+tabs2[1,2])/sum(tabs2)
ecrbootmodelrad<-(tabs3[2,1]+tabs3[1,2])/sum(tabs3)
ecrbootmodeltan<-(tabs4[2,1]+tabs4[1,2])/sum(tabs4)
 eblin[i,j]<-ecrbootmodellin
 ebpoly[i,j]<-ecrbootmodelpoly
 ebrad[i,j]<-ecrbootmodelrad
ebtan[i,j]<-ecrbootmodeltan
}
}