Stroke remains one of the leading causes of death and long-term disability worldwide. Early detection of stroke risk can be life-saving and with the power of machine learning (ML), we can now analyze patient data to predict who might be at higher risk.In this article, we’ll learn, how to use R programming to build predictive models using the Healthcare Stroke Dataset. You’ll learn how to explore the data, prepare it for modeling, and compare different algorithms like Logistic Regression, Decision Tree, and Random Forest.
Task One: Import Data and Preprocessing
Before building any model, it’s essential to load, clean, and explore the data.
Step 1. Load Required Packages
We’ll use several powerful R packages like tidyverse for data wrangling, caret for model training, and randomForest for building ensemble models, among others.
# Load Required Libraries library(tidyverse) library(skimr) library(DataExplorer) library(GGally) library(caret) library(tidymodels) library(corrplot) library(ROSE) library(e1071) library(rpart) library(randomForest) library(pROC) library(yardstick) library(janitor) library(ggthemes) library(shiny) library(vetiver)
Step 2. Load the Dataset
After loading the data, all the missing values are removed, ensuring the data is ready for analysis .
# Load Stroke Dataset healthcare_dataset_stroke_data <- read.csv("healthcare-dataset-stroke-data.csv")
stroke_data <- read.csv("healthcare-dataset-stroke-data.csv", na.strings = c("N/A", "", "NA")) sum(is.na(stroke_data)) # Total number of missing values colSums(is.na(stroke_data)) # Missing values per column stroke_data <- na.omit(stroke_data) sum(is.na(stroke_data)) # Should return 0
Step 3 : Describe and explore the data
Before jumping into modeling, let’s understand what the data looks like.
library(summarytools) # View first few rows head(stroke_data) # Check structure of the dataset str(stroke_data) # Quick summary of missing values colSums(is.na(stroke_data)) # Display basic summary statistics summary(stroke_data) # Display dimensions (rows and columns) dim(stroke_data) # View column names names(stroke_data) # Quick statistical summary using summarytools dfSummary(stroke_data) # Generate a basic exploratory report create_report(stroke_data)
Task 2 : Build prediction Model
Once data is ready, its time to build and test our model
Step 1. Data preparation for modelling
Categorical variable are need to be converted into factors, we will split data into training and testing subset.
library(caret)
# Check and convert categorical variables to factors
stroke_data$gender <- as.factor(stroke_data$gender)
stroke_data$ever_married <- as.factor(stroke_data$ever_married)
stroke_data$work_type <- as.factor(stroke_data$work_type)
stroke_data$Residence_type <- as.factor(stroke_data$Residence_type)
stroke_data$smoking_status <- as.factor(stroke_data$smoking_status)
stroke_data$stroke <- as.factor(stroke_data$stroke)
# Split dataset into training (70%) and testing (30%)
set.seed(123)
trainIndex <- createDataPartition(stroke_data$stroke, p = 0.7, list = FALSE)
train_data <- stroke_data[trainIndex, ]
test_data <- stroke_data[-trainIndex, ]
Model 1. Logistic Regression
Logistic Regression provides a baseline model that help us to understand which variables significantly influence stroke risk.
# 🧩 Model 1: Logistic Regression
#------------------------------------------
log_model <- glm(stroke ~ ., data = train_data, family = binomial)
summary(log_model)
# Predict on test data
log_pred <- predict(log_model, newdata = test_data, type = "response")
log_pred_class <- ifelse(log_pred > 0.5, 1, 0)
# Confusion matrix
confusionMatrix(as.factor(log_pred_class), test_data$stroke, positive = "1")
# Logistic regression
log_model <- glm(stroke ~ ., data = train_data, family = binomial)
# Predict
log_pred <- predict(log_model, newdata = test_data, type = "response")
log_pred_class <- ifelse(log_pred > 0.5, 1, 0)
# Confusion matrix
confusionMatrix(as.factor(log_pred_class), test_data$stroke, positive = "1")
Model 2. Random Forest
Here we train random forest model: to know the accuracy and robustness.
# 🌳 Model 2: Random Forest
#------------------------------------------
rf_model <- randomForest(stroke ~ ., data = train_data, ntree = 500, importance = TRUE)
rf_pred <- predict(rf_model, newdata = test_data)
# Evaluate performance
confusionMatrix(rf_pred, test_data$stroke, positive = "1")
# Variable importance plot
varImpPlot(rf_model)
Model 3. Decision Tree
It helps to visualize how risk factor splits.b
# 🌿 Model 3: Decision Tree #------------------------------------------ library(rpart) library(rpart.plot) library(caret) # Build the decision tree tree_model <- rpart(stroke ~ ., data = train_data, method = "class") # Visualize the tree rpart.plot(tree_model) # Make predictions on test data tree_pred <- predict(tree_model, newdata = test_data, type = "class") # Evaluate model performance confusionMatrix(tree_pred, test_data$stroke, positive = "1")
This specific project demonstrates the full workflow of a health data
1) Data import and preprocessing
2) Exploratory data analysis
3) Building and evaluating multiple prediction models
By Leveraging machine learning model, healthcare organizations can identify high- risk patient early, personalize preventive strategies and ultimately save lives.
