Random Forest (Classification)

Following post shows an overview of Random Forests using the Wisconsin Breast Cancer Dataset, from UCI Machine Learning Repository. Random Forests are a type of ensemble learning methods that learn from training number of individual decision trees, thereby reducing model variability and improving performance.
Wisconsin Breast Cancer Data, collected from the University of Wisconsin Hospitals, Madison from Dr. William H. Wolberg, is composed of 11 cytological attributes computed from digitized images of a fine needle aspirate (FNA) of a breast mass. Collected cell attributes include clump thickness, uniformity of cell size and cell shape, marginal adhesion, single epithelial cell size, bare nuclei, bland chromatin, normal nucleoli and mitoses. Features are used to differentiate between benign and malignant samples, as defined by the Class predictor. For further details about feature definition and data collection, see publication.


Import packages

library(randomForest)
library(caret)


Reading and Cleaning Wisconsin Breast Cancer Dataset from UCI Machine Learning Repository

breast_cancer <- read.csv("https://raw.githubusercontent.com/azkajavaid/BreastCancerWisconsinData-UCI/master/BreastCancerData.txt?token=ANkFlyu6ncAg-xoOxqhgB6wSw0GRnuBOks5atcg2wA%3D%3D")
colnames(breast_cancer) <- c("Code", "ClumpThickness", "UniformCellSize", "UniformCellShape", "MarginalAdhesion", "EpithelialCellSize", "BareNuclei", "BlandChromatin", "NormalNucleoli", "Mitoses", "Class")
breast_cancer$Class <- as.character(breast_cancer$Class)
breast_cancer$Class <- revalue(breast_cancer$Class, c("2" = "Benign", "4" = "Malignant"))
breast_cancer$Class <- as.factor(breast_cancer$Class)
breast_cancer <- breast_cancer[!(breast_cancer$BareNuclei == "?"),] # dropping all observations with BareNuclei value of "?"
breast_cancer$BareNuclei <- as.integer(breast_cancer$BareNuclei)


Split data in training and test sets

set.seed(90)
n <- nrow(breast_cancer)
index = sample(1:n, size = round(0.75*n), replace = FALSE)
train = breast_cancer[index, ]
test = breast_cancer[-index, ]
paste("Observations in training data: ", nrow(train), sep = "")

## [1] "Observations in training data: 512"

paste("Observations in testing data: ", nrow(test), sep = "")

## [1] "Observations in testing data: 170"


Fit Random Forest model, predicting Class

forest_model <- randomForest(Class ~ ., data = train)
forest_model

## 
## Call:
##  randomForest(formula = Class ~ ., data = train) 
##                Type of random forest: classification
##                      Number of trees: 500
## No. of variables tried at each split: 3
## 
##         OOB estimate of  error rate: 2.93%
## Confusion matrix:
##           Benign Malignant class.error
## Benign       320        10  0.03030303
## Malignant      5       177  0.02747253


Calculate variable importance

varImp(forest_model)

##                      Overall
## Code                4.990360
## ClumpThickness     12.905821
## UniformCellSize    53.571084
## UniformCellShape   57.252280
## MarginalAdhesion    6.969599
## EpithelialCellSize 28.332916
## BareNuclei         16.577294
## BlandChromatin     30.316033
## NormalNucleoli     21.874285
## Mitoses             1.278367

varImpPlot(forest_model)

"image"

Calculate accuracy for Class predictions on test data

predictions <- predict(forest_model, test)
forest_tab <- base::table(predictions, test$Class)
round(sum(diag(forest_tab))/sum(forest_tab), 3)

## [1] 0.976