random forest of classification

random forest
set.seed(311)
train.rf = train(useful ~ .,
                 data = Train,
                 method = "rf",
                 tuneGrid = data.frame(mtry = 1:120),
                 trControl = trainControl(method = "cv", number = 5, verboseIter = TRUE))
train.rf
train.rf$results

mod_rf = train.rf$finalModel
predict_rf = predict(mod_rf, newdata = Test)
table(Test$useful, predict_rf)


###predict with bootstrap

boot.accuracy.rf <- function(data,indices){
  data<-ggtm_complete[indices,]
  predict = predict(mod.boost, newdata = data, type = "class") 
  table<-table(data$useful, predict)
  accuracy<-sum(diag(table))/sum(table)
  return(accuracy)
  
}

boot.TPR.rf <- function(data,indices){
  data<-ggtm_complete[indices,]
  predict = predict(mod.boost, newdata = data, type = "class") 
  table<-table(data$useful, predict)
  TPR<-table[2,2]/(table[2,1]+table[2,2])
  return(TPR)
  
}

boot.FPR.rf <- function(data,indices){
  data<-ggtm_complete[indices,]
  predict = predict(mod.boost, newdata = data, type = "class") 
  table<-table(data$useful, predict)
  FPR<-table[1,2]/(table[1,1]+table[1,2])
  return(FPR)
  
}


boot.accuracy.rf.confi <-boot(data=Test, statistic = boot.accuracy.rf, R=100000)
boot.accuracy.rf.confi

boot.TPR.rf.confi <-boot(data=Test, statistic = boot.TPR.rf, R=100000)
boot.TPR.rf.confi

boot.FPR.rf.confi <-boot(data=Test, statistic = boot.FPR.rf, R=100000)
boot.FPR.rf.confi

boot.ci(boot.accuracy.rf.confi, conf = 0.95,type = "basic")
boot.ci(boot.TPR.rf.confi, conf = 0.95,type = "basic")
boot.ci(boot.FPR.rf.confi, conf = 0.95,type = "basic")

###problem c
Log = glm(useful ~ ., data = Train, family = "binomial")

summary(Log)

# Predictions on test set
PredictLog = predict(Log, newdata = Test, type = "response")
table(Test$useful, PredictLog > 0.6)
table<-table(Test$useful, PredictLog)
TPR<-table[2,2]/(table[2,1]+table[2,2])
FPR<-table[1,2]/(table[1,1]+table[1,2])
TPR
FPR