درخت تصمیم و جنگل تصادفی در R — راهنمای کاربردی

1#Set Directory and define response variable
2setwd("C:/Users/michaeljgrogan/Documents/a_documents/computing/data science/datasets")
3fullData <- read.csv("cars.csv")
4attach(fullData)
5
6fullData$response[CarSales > 24000] <- ">24000"
7fullData$response[CarSales > 1000 & CarSales <= 24000] <- ">1000 & <24000"
8fullData$response[CarSales <= 1000] <- "<1000"
9fullData$response<-as.factor(fullData$response)
10str(fullData)

1#Create training and test data
2inputData <- fullData[1:770, ] # training data
3testData <- fullData[771:963, ] # test data

1#Classification Tree
2library(rpart)
3formula=response~Age+Gender+Miles+Debt+Income
4dtree=rpart(formula,data=inputData,method="class",control=rpart.control(minsplit=30,cp=0.001))
5plot(dtree)
6text(dtree)
7summary(dtree)
8printcp(dtree)
9plotcp(dtree)
10printcp(dtree)

1> summary(dtree)
2
3Call:
4rpart(formula = formula, data = inputData, method = "class", 
5    control = rpart.control(minsplit = 30, cp = 0.001))
6  n= 770 
7
8           CP nsplit
91 0.496598639      0
102 0.013605442      1
113 0.008503401      6
124 0.001000000     10
13  rel error    xerror
141 1.0000000 1.0000000
152 0.5034014 0.5170068
163 0.4353741 0.5646259
174 0.4013605 0.5442177
18        xstd
191 0.07418908
202 0.05630200
213 0.05854027
224 0.05759793

1> #Prune the Tree and Plot
2pdtree<- prune(dtree, cp=dtree$cptable[which.min(dtree$cptable[,"xerror"]),"CP"])
3plot(pdtree, uniform=TRUE,
4     main="Pruned Classification Tree For Sales")
5text(pdtree, use.n=TRUE, all=TRUE, cex=.8)

1> #Model Testing
2> out <- predict(pdtree)
3> table(out[1:193],testData$response)
4> response_predicted <- colnames(out)[max.col(out, ties.method = c("first"))] # predicted
5> response_input <- as.character (testData$response) # actuals
6> mean (response_input != response_predicted) # misclassification %
7[1] 0.2844156

1> #Regression Tree
2fitreg <- rpart(CarSales~Age+Gender+Miles+Debt+Income,
3                method="anova", data=inputData)
4
5printcp(fitreg)
6plotcp(fitreg)
7summary(fitreg) 
8par(mfrow=c(1,2)) 
9rsq.rpart(fitreg) # cross-validation results

1> #Regression Tree
2> fitreg <- rpart(CarSales~Age+Gender+Miles+Debt+Income,
3+                 method="anova", data=inputData)
4> 
5> printcp(fitreg)
6
7Regression tree:
8rpart(formula = CarSales ~ Age + Gender + Miles + Debt + Income, 
9    data = inputData, method = "anova")
10
11Variables actually used in tree construction:
12[1] Age    Debt   Income
13
14Root node error: 6.283e+10/770 = 81597576
15
16n= 770 
17
18        CP nsplit rel error
191 0.698021      0   1.00000
202 0.094038      1   0.30198
213 0.028161      2   0.20794
224 0.023332      4   0.15162
235 0.010000      5   0.12829
24   xerror     xstd
251 1.00162 0.033055
262 0.30373 0.016490
273 0.21261 0.012890
284 0.18149 0.013298
295 0.14781 0.013068
30
31> plotcp(fitreg)
32> summary(fitreg)
33 
34Call:
35rpart(formula = CarSales ~ Age + Gender + Miles + Debt + Income, 
36    data = inputData, method = "anova")
37  n= 770 
38
39          CP nsplit rel error
401 0.69802077      0 1.0000000
412 0.09403824      1 0.3019792
423 0.02816107      2 0.2079410
434 0.02333197      4 0.1516189
445 0.01000000      5 0.1282869
45     xerror       xstd
461 1.0016159 0.03305536
472 0.3037301 0.01649002
483 0.2126110 0.01289041
494 0.1814939 0.01329778
505 0.1478078 0.01306756
51
52Variable importance
53  Debt  Miles Income    Age 
54    53     23     20      4

1> #Prune the Tree
2pfitreg<- prune(fitreg, cp=fitreg$cptable[which.min(fitreg$cptable[,"xerror"]),"CP"]) # from cptable   
3plot(pfitreg, uniform=TRUE,
4     main="Pruned Regression Tree for Sales")
5text(pfitreg, use.n=TRUE, all=TRUE, https://blog.faradars.org/wp-admin/post-new.php#cex=.8)

1> library(randomForest)
2> fitregforest <- randomForest(CarSales~Age+Gender+Miles+Debt+Income,data=inputData)
3> print(fitregforest) # view results
4
5Call:
6 randomForest(formula = CarSales ~ Age + Gender + Miles + Debt +      Income, data = inputData) 
7               Type of random forest: regression
8                     Number of trees: 500
9No. of variables tried at each split: 1
10
11          Mean of squared residuals: 10341022
12                    % Var explained: 87.33
13> importance(fitregforest) # importance of each predictor
14       IncNodePurity
15Age       5920357954
16Gender     187391341
17Miles    10811341575
18Debt     21813952812
19Income   12694331712