Random Forest in R

Random Forest:

· Ensemble of decision trees

· In decision tree, we have only one tree ,based on that we classify objects

· Whereas random forest generate multiple trees and take the decisions based on ensemble of all these trees

· More consistency

· For example, 500 models will be generated and classification can be done based on all these models

· For discrete attributes, class will be taken as majority predicted value (Classification Tree)

· For continuous attributes, class will be taken as average of all the predicted values (Regression Tree)

· Improves predictive accuracy

· Generates large number of boot strapped trees

· Final predicted outcome by combining the results across all of the trees

R Code:

>library(randomForest)

>Iris <- read.csv("E:/Dataset/Iris.csv") #Loading Training Data

>mydata=Iris #Moving Iris data set in to mydata

>mymodel=randomForest(Class ~ sepal.length+sepal.width+petal.length+petal.width,data=mydata) #model preparation

>mymodel

Call:

randomForest(formula = Class ~ sepal.length + sepal.width + petal.length + petal.width, data = mydata)

Type of random forest: classification

Number of trees: 500

No. of variables tried at each split: 2

OOB estimate of error rate: 4.67%

Confusion matrix:

Iris-setosa Iris-versicolor Iris-virginica class.error

Iris-setosa 50 0 0 0.00

Iris-versicolor 0 47 3 0.06

Iris-virginica 0 4 46 0.08

Interpretation:

1. 50 out of 50 are classified as Iris-setosa. i.e, 0.00 classification error
2. 47 out of 50 are classified as Iris-versicolor and 3 are misclassified as Iris-virginica. i.e, 0.06 classification error
3. 46 out of 50 are classified as Iris-virginica and 4 are misclassified as Iris-versicolor. i.e, 0.08 classification error

>newdata=read.csv("E:/Dataset/Iris_test.csv") #Loading Test Data

>view(newdata)

>pred=predict(mymodel,newdata=newdata) #validation

>mytable=table(newdata$Class,pred) #confusion matrix

>mytable

pred

Iris-setosa Iris-versicolor Iris-virginica

Iris-setosa 49 0 0

Iris-versicolor 0 15 0

Iris-virginica 0 0 2

Interpretation:

1. 49 out of 49 are classified into Iris-setosa. i.e, 100 % Accuracy.
2. 15 out of 15 are classified into Iris-versicolor. i.e,100 % Accuracy.
3. 2 out of 2 are classified into Iris-virginica. i.e, 100% Accuracy
4. none of the object is misclassified in the above model for the given test data.