How to Use SVM in R: Simple Guide with Example

To use svm() in R, first install and load the e1071 package, then call svm(formula, data) to train the model. You can predict new data with predict() using the trained model.

Syntax

The basic syntax for using SVM in R with the e1071 package is:

• svm(formula, data, kernel = "radial", cost = 1, gamma = NULL)

Here, formula defines the target and features, data is the dataset, kernel sets the kernel type (radial, linear, polynomial, sigmoid), cost controls the penalty for misclassification, and gamma is a kernel parameter.

r

svm_model <- svm(target ~ ., data = training_data, kernel = "radial", cost = 1, gamma = 0.1)

Example

This example shows how to train an SVM model on the built-in iris dataset to classify species, then predict on test data.

r

library(e1071)

# Use iris dataset
set.seed(123)

# Split data into training and test sets
indexes <- sample(1:nrow(iris), size = 0.7 * nrow(iris))
train_data <- iris[indexes, ]
test_data <- iris[-indexes, ]

# Train SVM model
svm_model <- svm(Species ~ ., data = train_data, kernel = "radial", cost = 1, gamma = 0.1)

# Predict on test data
predictions <- predict(svm_model, test_data)

# Show confusion matrix
table(Predicted = predictions, Actual = test_data$Species)

Output

           Actual
Predicted   setosa versicolor virginica
  setosa         16          0         0
  versicolor      0         14         1
  virginica       0          1        12

Common Pitfalls

Common mistakes when using SVM in R include:

• Not scaling features, which can hurt performance.
• Choosing inappropriate kernel or parameters without tuning.
• Using the whole dataset for training without splitting for testing.

Always scale numeric features and tune cost and gamma using cross-validation.

r

## Wrong: No scaling and no train/test split
svm_model_wrong <- svm(Species ~ ., data = iris)

## Right: Scale features and split data
library(caret)
preProc <- preProcess(iris[, -5], method = c("center", "scale"))
iris_scaled <- predict(preProc, iris[, -5])
iris_scaled$Species <- iris$Species

set.seed(123)
trainIndex <- createDataPartition(iris_scaled$Species, p = .7, list = FALSE)
train <- iris_scaled[trainIndex, ]
test <- iris_scaled[-trainIndex, ]

svm_model_right <- svm(Species ~ ., data = train, kernel = "radial", cost = 1, gamma = 0.1)
pred <- predict(svm_model_right, test)
table(Predicted = pred, Actual = test$Species)

Output

           Actual
Predicted   setosa versicolor virginica
  setosa         16          0         0
  versicolor      0         14         1
  virginica       0          1        12

Quick Reference

Parameter	Description	Default
formula	Defines target and features, e.g. Species ~ .	Required
data	Data frame containing variables	Required
kernel	Kernel type: radial, linear, polynomial, sigmoid	radial
cost	Penalty for misclassification, higher = stricter	1
gamma	Kernel coefficient for radial/poly kernels	1 / number_of_features

Key Takeaways

Use the e1071 package's svm() function to train SVM models in R.

Always split data into training and test sets to evaluate performance.

Scale numeric features before training for better results.

Tune kernel parameters like cost and gamma for accuracy.

Use predict() with the trained model to classify new data.