install.packages("randomForest")

#只有caret包才有混淆矩陣

#set.seed保證每次運(yùn)行代碼獲得相同的隨機(jī)樣本

npdata <- read.table("data.txt",header=T)

set.seed(123)

#抽取訓(xùn)練集和測試集

train_sub <- sample(nrow(npdata),0.7*nrow(npdata ))#隨機(jī)無放回抽取0.7樣本

train_data <- npdata[train_sub,]

test_data <- npdata[-train_sub,]

library(randomForest)

#導(dǎo)入數(shù)據(jù)

#隨機(jī)森林計(jì)算

set.seed(100)

np.rf <- randomForest(npl~.,data=train_data,importance=TRUE,ntree=50)

#使用訓(xùn)練集查看預(yù)測精度

np_predict <- predict(np.rf,train_data)

plot(train_data$npl,np_predict,main='訓(xùn)練集', xlab = 'npl',ylab = 'Predict')

abline(1,1)

#使用測試集評估預(yù)測性能

np_predict <- predict(np.rf,test_data)

plot(test_data$npl,np_predict,main='預(yù)測集', xlab = 'npl',ylab = 'Predict')

abline(1,1)

#查看擬合優(yōu)度

print(np.rf)

importance(np.rf)

varImpPlot(np.rf,main = "variable importance")

#最優(yōu)決策樹數(shù)量

plot(np.rf)

#繪制錯(cuò)誤率表,確定變量個(gè)數(shù)

result = rfcv(train_data[-1],train_data$npl,cv.fold = 10)

result$error.cv

with(result,plot(n.var,error.cv,log="x",type = "o",lwd=2))