Support Vector Machines for Heart Failure Prediction.Rmd

---
title: "TM"
author: "MSHTSA009"
date: "2024-05-24"
output: pdf_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```


```{r cars}
#a Step a: Randomly split the data into training and testing sets by applying an 80-20 split
# Load the necessary libraries
library(e1071)
library(caret)
library(pROC)

# Load the data
data <- read.csv("C:/Users/ellio/Downloads/heart_failure_clinical_records_dataset.csv")
# Convert DEATH_EVENT to a factor
data$DEATH_EVENT <- as.factor(data$DEATH_EVENT)
print(levels(data$DEATH_EVENT))  # Check the levels of DEATH_EVENT

# Split the data into training and testing sets
set.seed(1)
trainIndex <- createDataPartition(data$DEATH_EVENT, p = 0.8, list = FALSE)
trainData <- data[trainIndex, ]
testData <- data[-trainIndex, ]

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#b Step b: Build a support vector machine using a radial kernel function with cost=0.1 and gamma=0.1. Report the classification accuracy, recall, specificity, F1 score, and ROC AUC.

# Train SVM with radial kernel
svm_model <- svm(DEATH_EVENT ~ ., data = trainData, kernel = "radial", cost = 0.1, gamma = 0.1, probability = TRUE)

# Predict on the test data
pred <- predict(svm_model, testData, probability = TRUE)
pred_prob <- attr(pred, "probabilities")[, 2]

# Convert predictions to factors with the same levels as DEATH_EVENT
pred <- factor(pred, levels = levels(testData$DEATH_EVENT))

# Calculate metrics
conf_matrix <- confusionMatrix(pred, testData$DEATH_EVENT)
accuracy <- conf_matrix$overall['Accuracy']
recall <- conf_matrix$byClass['Recall']
specificity <- conf_matrix$byClass['Specificity']
f1 <- conf_matrix$byClass['F1']
roc_auc <- roc(testData$DEATH_EVENT, pred_prob)$auc

# Print the metrics
print(paste("Accuracy:", accuracy))
print(paste("Recall:", recall))
print(paste("Specificity:", specificity))
print(paste("F1 Score:", f1))
print(paste("ROC AUC:", roc_auc))

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
# Step c: Repeat steps (a) and (b) 100 times and report the average of each of these metrics over the 100 runs. Also, provide a boxplot for each metric’s 100 values.
# Repeat the process 100 times
results <- data.frame(Accuracy = numeric(100), Recall = numeric(100), Specificity = numeric(100), F1 = numeric(100), ROC_AUC = numeric(100))

for (i in 1:100) {
  set.seed(i)
  trainIndex <- createDataPartition(data$DEATH_EVENT, p = 0.8, list = FALSE)
  trainData <- data[trainIndex, ]
  testData <- data[-trainIndex, ]
  
  svm_model <- svm(DEATH_EVENT ~ ., data = trainData, kernel = "radial", cost = 0.1, gamma = 0.1, probability = TRUE)
  pred <- predict(svm_model, testData, probability = TRUE)
  pred_prob <- attr(pred, "probabilities")[, 2]
  
  pred <- factor(pred, levels = levels(testData$DEATH_EVENT))
  
  conf_matrix <- confusionMatrix(pred, testData$DEATH_EVENT)
  results$Accuracy[i] <- conf_matrix$overall['Accuracy']
  results$Recall[i] <- conf_matrix$byClass['Recall']
  results$Specificity[i] <- conf_matrix$byClass['Specificity']
  results$F1[i] <- conf_matrix$byClass['F1']
  results$ROC_AUC[i] <- roc(testData$DEATH_EVENT, pred_prob)$auc
}

# Report the average of each metric over the 100 runs
avg_metrics <- colMeans(results)
print(avg_metrics)

# Provide a boxplot for each metric’s 100 values
boxplot(results, main="SVM Metrics Distribution", xlab="Metrics", ylab="Values")

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#Step d: Repeat steps (a), (b) and (c) with different cost and gamma parameters. You are expected to explore at least 3 different values for each hyperparameter. Report your findings using appropriate tables and/or graphs.

# Repeat the process with different cost and gamma parameters
cost_values <- c(0.01, 0.1, 1)
gamma_values <- c(0.01, 0.1, 1)
grid_results <- expand.grid(Cost = cost_values, Gamma = gamma_values, Accuracy = numeric(0), Recall = numeric(0), Specificity = numeric(0), F1 = numeric(0), ROC_AUC = numeric(0))

for (cost in cost_values) {
  for (gamma in gamma_values) {
    for (i in 1:100) {
      set.seed(i)
      trainIndex <- createDataPartition(data$DEATH_EVENT, p = 0.8, list = FALSE)
      trainData <- data[trainIndex, ]
      testData <- data[-trainIndex, ]
      
      svm_model <- svm(DEATH_EVENT ~ ., data = trainData, kernel = "radial", cost = cost, gamma = gamma, probability = TRUE)
      pred <- predict(svm_model, testData, probability = TRUE)
      pred_prob <- attr(pred, "probabilities")[, 2]
      
      pred <- factor(pred, levels = levels(testData$DEATH_EVENT))
      
      conf_matrix <- confusionMatrix(pred, testData$DEATH_EVENT)
      grid_results <- rbind(grid_results, data.frame(Cost = cost, Gamma = gamma, Accuracy = conf_matrix$overall['Accuracy'], Recall = conf_matrix$byClass['Recall'], Specificity = conf_matrix$byClass['Specificity'], F1 = conf_matrix$byClass['F1'], ROC_AUC = roc(testData$DEATH_EVENT, pred_prob)$auc))
    }
  }
}


# Summarize the results
summary_results <- aggregate(. ~ Cost + Gamma, data = grid_results, mean)
print(summary_results)

#Plot results
library(ggplot2)

ggplot(summary_results, aes(x = Cost, y = Accuracy, color = factor(Gamma))) +
  geom_line(size = 1.5) +     # Adjust line width here
  geom_point(size = 3) +      # Adjust point size here
  labs(title = "", 
       x = "Cost", y = "Accuracy",
       size = 19) +             # Adjust text size here
  scale_color_discrete(name = "Gamma")


#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


#Step e: Tune the cost and gamma parameters using grid search on the training data and report your best model parameters. Test your best model performance on the test data.

# Tune the cost and gamma parameters using grid search
set.seed(1)
tune_result <- tune(svm, DEATH_EVENT ~ ., data = trainData, kernel = "radial", ranges = list(cost = cost_values, gamma = gamma_values), probability = TRUE)

# Report the best model parameters
best_model <- tune_result$best.model
best_parameters <- tune_result$best.parameters
print(best_parameters)


# Test the best model performance on the test data
pred_best <- predict(best_model, testData, probability = TRUE)
pred_prob_best <- attr(pred_best, "probabilities")[, 2]

# Convert predictions to factors with the same levels as DEATH_EVENT
pred_best <- factor(pred_best, levels = levels(testData$DEATH_EVENT))

# Calculate metrics for the best model
conf_matrix_best <- confusionMatrix(pred_best, testData$DEATH_EVENT)
accuracy_best <- conf_matrix_best$overall['Accuracy']
recall_best <- conf_matrix_best$byClass['Recall']
specificity_best <- conf_matrix_best$byClass['Specificity']
f1_best <- conf_matrix_best$byClass['F1']
roc_auc_best <- roc(testData$DEATH_EVENT, pred_prob_best)$auc

# Print the metrics for the best model
print(paste("Best Model Accuracy:", accuracy_best))
print(paste("Best Model Recall:", recall_best))
print(paste("Best Model Specificity:", specificity_best))
print(paste("Best Model F1 Score:", f1_best))
print(paste("Best Model ROC AUC:", roc_auc_best))

#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#Step f: Compare findings from the grid search and repeated runscomparison <- data.frame(
# Compare findings from the grid search and repeated runs
comparison <- data.frame(
  Metric = c("Accuracy", "Recall", "Specificity", "F1", "ROC_AUC"),
  Grid_Search = c(accuracy_best, recall_best, specificity_best, f1_best, roc_auc_best), 
  Repeated_Runs = avg_metrics)
print(comparison)
```