Question

- For the spam data, partition the data into 2/3 training and 1/3 test data.

- Find the best 12 variables whose t-test statistics (in absolute value) are highest 12.

#You may use apply function to get 12 variable names.

- Build the GAM model for spam training data using the first 12 variables whose t.test (two sample) statistic (in absolute value) are within top 12.

r coding?

Answer 1

Given that

Will only provide R codes

##a1: splitting the data##

require(kernlab)
data=spam
select=sample(1:nrow(spam),ceiling(nrow(spam)/3),replace=F)
test=data[select,]
train=data[-select,]

##a2: Placing an initial model and finding the best 12 variables##
attach(train)

l=glm(type~.,data=train,family ="binomial")

best=sort.int(summary(l)$coeff[,4][-1],dec=F,index.return=T)$ix[1:12] # This gives the indices of best 12 variables

##b: FInal GLm model ##

train_final=cbind(train[,best],type)
final = glm(type~.,data=train_final,family="binomial")