R Machine Learning with caret — Predictive Modeling

Learning Objectives

By the end of this tutorial, you will be able to:

• Prepare data for machine learning (splitting, preprocessing)
• Train and tune models using the caret package
• Evaluate model performance with cross-validation
• Compare multiple algorithms
• Build prediction pipelines

The caret Package

library(caret)

# Check available algorithms
modelLookup()

Data Preparation

# Train/test split
set.seed(42)
train_index <- createDataPartition(iris$Species, p = 0.7, list = FALSE)
train_data <- iris[train_index, ]
test_data <- iris[-train_index, ]

# Cross-validation
train_control <- trainControl(
  method = "cv",
  number = 10,
  classProbs = TRUE,
  summaryFunction = twoClassSummary
)

# Preprocessing
preprocess_params <- preProcess(train_data[, -5], method = c("center", "scale"))
train_scaled <- predict(preprocess_params, train_data[, -5])

Classification

Logistic Regression

model_glm <- train(
  Species ~ .,
  data = iris,
  method = "glm",
  trControl = train_control,
  metric = "Accuracy"
)
print(model_glm)

Random Forest

model_rf <- train(
  Species ~ .,
  data = iris,
  method = "rf",
  trControl = train_control,
  tuneGrid = expand.grid(mtry = c(1, 2, 3, 4))
)
print(model_rf)
plot(model_rf)

Support Vector Machine

model_svm <- train(
  Species ~ .,
  data = iris,
  method = "svmRadial",
  trControl = train_control,
  preProcess = c("center", "scale")
)
print(model_svm)

K-Nearest Neighbors

model_knn <- train(
  Species ~ .,
  data = iris,
  method = "knn",
  trControl = train_control,
  preProcess = c("center", "scale"),
  tuneGrid = expand.grid(k = seq(1, 21, by = 2))
)
print(model_knn)
plot(model_knn)

Regression

# Random Forest for regression
model_rf_reg <- train(
  mpg ~ .,
  data = mtcars,
  method = "rf",
  trControl = trainControl(method = "cv", number = 10),
  tuneGrid = expand.grid(mtry = c(2, 4, 6, 8))
)
print(model_rf_reg)

# Predictions
predictions <- predict(model_rf_reg, newdata = mtcars)
postResample(pred = predictions, obs = mtcars$mpg)

Model Comparison

# Train multiple models
models <- resamples(list(
  GLM = model_glm,
  RF = model_rf,
  SVM = model_svm,
  KNN = model_knn
))

# Summary
summary(models)

# Box plots
bwplot(models, metric = "Accuracy")

# Dot plots
dotplot(models, metric = "Accuracy")

Feature Selection

# Recursive Feature Elimination
rfe_control <- rfeControl(functions = rfFuncs, method = "cv", number = 10)

rfe_result <- rfe(
  iris[, -5],
  iris$Species,
  sizes = c(1, 2, 3, 4),
  rfeControl = rfe_control
)
print(rfe_result)
plot(rfe_result, type = c("g", "o"))

# Variable importance
varImp(model_rf)
plot(varImp(model_rf))

Practical Examples

Example 1: Customer Churn Prediction

library(caret)

# Simulated data
set.seed(42)
n <- 1000
data <- data.frame(
  tenure = sample(1:72, n, replace = TRUE),
  monthly_charges = runif(n, 20, 100),
  contract = factor(sample(c("Month-to-month", "One year", "Two year"), n, replace = TRUE))
)
data$churn <- factor(ifelse(
  data$contract == "Month-to-month" & data$monthly_charges > 60,
  sample(c("Yes", "No"), n, replace = TRUE, prob = c(0.4, 0.6)),
  sample(c("Yes", "No"), n, replace = TRUE, prob = c(0.1, 0.9))
))

# Split
train_index <- createDataPartition(data$churn, p = 0.7, list = FALSE)
train <- data[train_index, ]
test <- data[-train_index, ]

# Train model
model <- train(
  churn ~ .,
  data = train,
  method = "rf",
  trControl = trainControl(method = "cv", number = 5),
  tuneGrid = expand.grid(mtry = c(1, 2, 3))
)

# Evaluate
predictions <- predict(model, newdata = test)
confusionMatrix(predictions, test$churn)

Practice Exercises

Exercise 1: Model Comparison

Compare logistic regression, random forest, and SVM on the Sonar dataset.

Solution

library(caret)
library(mlbench)
data(Sonar)

set.seed(42)
train_index <- createDataPartition(Sonar$Class, p = 0.7, list = FALSE)
train <- Sonar[train_index, ]
test <- Sonar[-train_index, ]

ctrl <- trainControl(method = "cv", number = 5)

model_glm <- train(Class ~ ., data = train, method = "glm", trControl = ctrl)
model_rf <- train(Class ~ ., data = train, method = "rf", trControl = ctrl)
model_svm <- train(Class ~ ., data = train, method = "svmRadial", trControl = ctrl)

# Compare
resamps <- resamples(list(GLM = model_glm, RF = model_rf, SVM = model_svm))
summary(resamps)
bwplot(resamps)

Key Takeaways

• caret provides a unified interface for hundreds of algorithms
• trainControl() defines resampling methods (CV, bootstrap)
• train() fits models with automatic hyperparameter tuning
• confusionMatrix() evaluates classification performance
• postResample() evaluates regression performance
• resamples() compares multiple models
• varImp() shows feature importance
• Always preprocess — center, scale, handle missing values

Next: Learn about R Shiny Web Apps — interactive web applications.