Class Example – Supervised Learning

Consider the Breast Cancer dataset (BreastCancer.csv). You would like to model and predict if a given
specimen is benign or malignant, based on the other cell features present in the data.

Import and include packages

In the cell below, install and include all necessary packages

library(caret)

## Loading required package: lattice

## Loading required package: ggplot2

Import the necessary data

In the cell below, import the necessary dataset(s)

bc <- read.csv("BreastCancer.csv")
head(bc)

## X Cl.thickness Cell.size Cell.shape Marg.adhesion Epith.c.size Bare.nuclei

## 1 1 5 1 1 1 2 1
## 2 2 5 4 4 5 7 10
## 3 3 3 1 1 1 2 2
## 4 4 6 8 8 1 3 4
## 5 5 4 1 1 3 2 1
## 6 6 8 10 10 8 7 10
## Bl.cromatin Normal.nucleoli Mitoses Class
## 1 3 1 1 benign
## 2 3 2 1 benign
## 3 3 1 1 benign
## 4 3 7 1 benign
## 5 3 1 1 benign
## 6 9 7 1 malignant

Step 1: Dummy Variables

• Response variable is a binary qualitative variable.

• These variables need to be represented numerically in R to make analysis possible.
• Dummy variables can be used as a numerical “stand-in” for qualitative variables.
• Typically, a dummy variable takes on the value of 0 or 1 to indicate the absence or presence of a factor.
• Define a dummy variable using the ifelse( ) function.

1
 • Convert variable type for new dummy variable from numeric to factor.

For this example, - Indicate a benign response with a 0 and a malignant response with a 1.

bc$Class <- ifelse(bc$Class == "benign", 0, 1)

#bc$Class <- ifelse(bc$Class == "malignant", 1, 0)
head(bc)

## X Cl.thickness Cell.size Cell.shape Marg.adhesion Epith.c.size Bare.nuclei

## 1 1 5 1 1 1 2 1
## 2 2 5 4 4 5 7 10
## 3 3 3 1 1 1 2 2
## 4 4 6 8 8 1 3 4
## 5 5 4 1 1 3 2 1
## 6 6 8 10 10 8 7 10
## Bl.cromatin Normal.nucleoli Mitoses Class
## 1 3 1 1 0
## 2 3 2 1 0
## 3 3 1 1 0
## 4 3 7 1 0
## 5 3 1 1 0
## 6 9 7 1 1

class(bc$Class)

## [1] "numeric"

bc$Class <- factor(bc$Class, levels = c(0,1))

head(bc$Class)

## [1] 0 0 0 0 0 1
## Levels: 0 1

Step 2: Train/Test Split

• Randomly split the data into a training and test set.

For this example,

• Randomly split the data into a training and test set using a 80/20% split.
• Use a seed value of 404.

set.seed(404)
split <- round(nrow(bc)*0.80)
train_ind <- sample(1:nrow(bc),
split,
replace = FALSE)
trainData <- bc[train_ind,]
testData <- bc[-train_ind,]
head(trainData)

2
## X Cl.thickness Cell.size Cell.shape Marg.adhesion Epith.c.size
## 441 456 10 2 2 1 2
## 536 551 3 1 1 1 2
## 570 585 5 1 1 6 3
## 39 40 2 5 3 3 6
## 552 567 3 1 2 1 2
## 109 111 1 3 1 2 2
## Bare.nuclei Bl.cromatin Normal.nucleoli Mitoses Class
## 441 6 1 1 2 1
## 536 1 2 1 1 0
## 570 1 1 1 1 0
## 39 7 7 5 1 1
## 552 1 3 1 1 0
## 109 2 5 3 2 0

Step 3: Fit the Logistic Regression Model

• Use the glm( ) function in R to fit a logistic regression model. The general syntax for this model is,

– glm(y~., data = my_dataset, family = ‘binomial’)

– glm(y~x1+x2+x3, data = my_dataset, family = ‘binomial’)

For this example, - Build a logistic regression model with Class being our response variable and the remaining
variables as predictors.

logistic_mod <- glm(Class ~ ., data = trainData, family = "binomial")

summary(logistic_mod)

##
## Call:
## glm(formula = Class ~ ., family = "binomial", data = trainData)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -3.3735 -0.0835 -0.0378 0.0207 2.0224
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -10.608439 1.697683 -6.249 4.14e-10 ***
## X -0.002028 0.001844 -1.100 0.271377
## Cl.thickness 0.698433 0.187110 3.733 0.000189 ***
## Cell.size 0.031699 0.234893 0.135 0.892651
## Cell.shape 0.317733 0.249407 1.274 0.202680
## Marg.adhesion 0.295125 0.145972 2.022 0.043198 *
## Epith.c.size 0.272023 0.197751 1.376 0.168951
## Bare.nuclei 0.385287 0.108562 3.549 0.000387 ***
## Bl.cromatin 0.442263 0.205410 2.153 0.031313 *
## Normal.nucleoli 0.107847 0.118764 0.908 0.363837
## Mitoses 0.245077 0.328227 0.747 0.455262
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)

3
##
## Null deviance: 703.096 on 545 degrees of freedom
## Residual deviance: 71.212 on 535 degrees of freedom
## AIC: 93.212
##
## Number of Fisher Scoring iterations: 8

logistic_model_ex <- glm(Class ~ Cl.thickness + Cell.size + Cell.shape,

data = trainData, family = "binomial")
summary(logistic_model_ex)

##
## Call:
## glm(formula = Class ~ Cl.thickness + Cell.size + Cell.shape,
## family = "binomial", data = trainData)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -3.8293 -0.1680 -0.0881 0.0128 2.2697
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -8.3155 0.8634 -9.631 < 2e-16 ***
## Cl.thickness 0.6481 0.1267 5.117 3.11e-07 ***
## Cell.size 0.7616 0.2147 3.547 0.000389 ***
## Cell.shape 0.7081 0.1968 3.597 0.000321 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 703.10 on 545 degrees of freedom
## Residual deviance: 130.05 on 542 degrees of freedom
## AIC: 138.05
##
## Number of Fisher Scoring iterations: 8

Step 4: Predictions

• Use the predict( ) function in R to make predictions based on the Logistic Regression model in Step 3.
• The general syntax for the predict( ) function is,
– predict(model, my_dataset, type = ‘response’)
• The predict( ) function returns probabilities.
• Therefore, we need to set a threshold value to classify each predicted outcome to a specific category.
• To classify the probabilities into two groups, use the ifelse( ) function in R.

For this example, - Predict the response variables for the test set.

# Predict the response variables for the test set

pred <- predict(logistic_mod, newdata = testData, type = "response")
head(pred)

4
## 9 13 14 15 24 26
## 0.002785780 0.256632907 0.002693492 0.999851238 0.001220712 0.002480340

# Recode factors
y_pred_num <- ifelse(pred > 0.5, 1, 0)
y_pred <- factor(y_pred_num) # convert the numeric predictors response to factors
y_act <- testData$Class

head(y_pred)

## 9 13 14 15 24 26
## 0 0 0 1 0 0
## Levels: 0 1

head(y_act)

## [1] 0 1 0 1 0 0
## Levels: 0 1

Step 5

• Use the confusionMatrix( ) function from the caret package to obtain the confusion matrix of the
observed and predicted classes.
• The general syntax for the confusionMatrix( ) function is,
• confusionMatrix(data, reference, positive = “positive_factor_level”)

For this example, - Calculate the performance metrics.

# Performance metrics
confusionMatrix(data = y_pred, reference = y_act, positive = "1")

## Confusion Matrix and Statistics

##
## Reference
## Prediction 0 1
## 0 85 5
## 1 1 46
##
## Accuracy : 0.9562
## 95% CI : (0.9071, 0.9838)
## No Information Rate : 0.6277
## P-Value [Acc > NIR] : <2e-16
##
## Kappa : 0.9048
##
## Mcnemar's Test P-Value : 0.2207
##
## Sensitivity : 0.9020
## Specificity : 0.9884
## Pos Pred Value : 0.9787
## Neg Pred Value : 0.9444

5
## Prevalence : 0.3723
## Detection Rate : 0.3358
## Detection Prevalence : 0.3431
## Balanced Accuracy : 0.9452
##
## 'Positive' Class : 1
##