回歸樹模型在之前的博客文章中已有介紹,。而裝袋算法與隨機(jī)森林相對而言會生成多個樹模型，再進(jìn)行組合預(yù)測,，其效果遠(yuǎn)大于單個樹模型,。裝袋算法（bagging）采取自助法的思路，從樣本中隨機(jī)抽樣,，形成多個訓(xùn)練樣本,，生成多個樹模型。然后以多數(shù)投票的方式來預(yù)測結(jié)果,。隨機(jī)森林則(randomForest)更進(jìn)一步,，不僅對樣本進(jìn)行抽樣，還對變量進(jìn)行抽樣,。下面來橫向?qū)Ρ纫幌赂魉惴ā?BR>
首先讀入必要的程序包
library(DMwR)
library(rpart)
library(ipred)
library(randomForest)

前二種算法可以計算缺失數(shù)據(jù),，但隨機(jī)森林不行，所以還需將數(shù)據(jù)進(jìn)行清洗整理

data(algae)
algae <- algae[-manyNAs(algae), ]
clean.algae <- knnImputation(algae,k=10)

回歸樹模型計算
model.tree=rpart(a1 ~ ., data = clean.algae[, 1:12])
pre.tree <- predict(model.tree, clean.algae)
plot(pre.tree~clean.algae$a1)
nmse1 <- mean((pre.tree- clean.algae[,'a1'])^2)/
mean((mean(clean.algae[,'a1'])- clean.algae[,'a1'])^2)

裝袋算法計算
model.bagging <- bagging(
a1 ~ ., data = clean.algae[, 1:12], nbagg=1000)
pre.bagging=predict(model.bagging,clean.algae)
plot(pre.bagging~clean.algae$a1)
nmse2 <- mean((pre.bagging- clean.algae[,'a1'])^2)/
mean((mean(clean.algae[,'a1'])- clean.algae[,'a1'])^2)

隨機(jī)森林計算
model.forest <-randomForest(a1 ~ ., data = clean.algae)
#若有缺失數(shù)據(jù)需加入： na.action=na.omit
pre.forest=predict(model.forest, clean.algae)
plot(pre.forest~ clean.algae$a1)
(nmse3 <- mean((pre.forest- clean.algae[,'a1'])^2)/
mean((mean( clean.algae[,'a1'])- clean.algae[,'a1'])^2)

print(c(nmse1,nmse2,nmse3))
用預(yù)測值與真值之間的相對離差平方和來作為測量誤差的指標(biāo),，其結(jié)果分別為：0.3541180 0.3103366 0.1002235 可以看出隨機(jī)森林是最有效的,。

再來看看處理分類數(shù)據(jù)的表現(xiàn)，利用iris數(shù)據(jù)來判斷花的種類

library(randomForest)
model.forest <-randomForest(Species ~ ., data = iris)
pre.forest=predict(model.forest, iris)
table(pre.forest,iris$Species)

pre.forest   setosa versicolor virginica
  setosa         50          0         0
  versicolor      0         50         0
  virginica       0          0        50

library(rpart)
model.tree=rpart(Species ~ ., data = iris,method='class')
pre.tree=predict(model.tree, data = iris,type='class')
table(pre.tree,iris$Species)

pre.tree     setosa versicolor virginica
  setosa         50          0         0
  versicolor      0         49         5
  virginica       0          1        45

隨機(jī)森林算法預(yù)測全對,，而分類樹模型則有六處錯誤,。