Random Forest on Titanic Dataset

# Install and load necessary packages
library(randomForest)
library(caret)
library(ggplot2)

# Load the Titanic dataset
data("Titanic")

# Convert the dataset to a data frame
titanic_df <- as.data.frame(Titanic)
head(titanic_df)

Convert ‘Survived’ to a binary factor

set.seed(123)
titanic_df$Survived <- as.factor(ifelse(titanic_df$Survived == "Yes", 1, 0))

splitIndex <- caret::createDataPartition(titanic_df$Survived, p = 0.7, list = FALSE)
train_data <- titanic_df[splitIndex, ]
test_data <- titanic_df[-splitIndex, ]

Split the data into training and testing datasets

set.seed(123)
splitIndex <- caret::createDataPartition(titanic_df$Survived, p = 0.7, list = FALSE)
train_data <- titanic_df[splitIndex, ]
test_data <- titanic_df[-splitIndex, ]

rf_model <- randomForest::randomForest(Survived ~ ., data = train_data, ntree = 100)

# Feature selection using importance measure from the random forest
importance <- randomForest::importance(rf_model)

# Extract feature importance
importance$Feature <- rownames(importance)

Warning: Coercing LHS to a list

# Select features with positive MeanDecreaseGini
selected_features <- importance$Feature[importance$MeanDecreaseGini > 0]

# Subset the data with selected features
train_data_selected <- train_data[, c("Survived", selected_features)]
test_data_selected <- test_data[, c("Survived", selected_features)]

# Extract feature importance
importance <- randomForest::importance(rf_model)
importance$Feature <- rownames(importance)

Warning: Coercing LHS to a list

# Select features with positive MeanDecreaseGini
selected_features <- importance$Feature[importance$MeanDecreaseGini > 0]

# Subset the data with selected features
train_data_selected <- train_data[, c("Survived", selected_features)]
test_data_selected <- test_data[, c("Survived", selected_features)]

top_n <- 4
# Visualize feature importance
# Check the structure of importance object
str(importance)

List of 5
 $        : num 1.79
 $        : num 0.666
 $        : num 1
 $        : num 6.92
 $ Feature: chr [1:4] "Class" "Sex" "Age" "Freq"

# Extract feature importance
importance_df <- data.frame(
  Feature = importance[[5]],
  Importance = unlist(importance[1:4])
)

# Order the data frame by Importance
importance_df <- importance_df[order(-importance_df$Importance), ]

# Keep only the top N features
importance_df <- importance_df[1:top_n, ]


importance_df <- importance_df[1:top_n, ]  # Keep only the top N features

Plot

# Plot the top N important features

# Visualize feature importance
print(ggplot2::ggplot(importance_df, ggplot2::aes(x = reorder(Feature, Importance), y = Importance)) +
      ggplot2::geom_bar(stat = "identity", fill = "skyblue", color = "black") +
      ggplot2::coord_flip() +
      ggplot2::labs(title = paste("Top", top_n, "Important Features from Random Forest"),
           x = "Feature",
           y = "Mean Decrease in Gini"))