R script.R

library(MASS)           
library(nnet)           
library(ggplot2)    
library(caret) 
install.packages("ggpubr")         
library(ggpubr)         

df <- read.csv("heart.csv")

# Extracting Numerical Variables
df <- df[, c("Age", "RestingBP", "Cholesterol", "MaxHR", "HeartDisease")]
df$HeartDisease <- as.factor(df$HeartDisease)  # Convert target to factor

# QQ Plots
qq_plots <- list()
for (var in colnames(df)[1:4]) {
  p <- ggqqplot(df[[var]], ylab = var) + ggtitle(paste("QQ Plot for", var))
  qq_plots[[var]] <- p
}
ggarrange(plotlist = qq_plots, ncol = 2, nrow = 2)

# Internal Validation
# Dr. Ali's Function
flda <- function(x, class) {
  cat("Fisher Linear Discriminant:\n")
  a <- lda(x, class)
  d <- predict(a)
  t <- table(class, d$class)
  print(t)
  er <- 100 * (sum(t) - sum(diag(t))) / nrow(x)
  cat("Error Rate =", er, "%\n")
  return(list(predictions = d, error_rate = er))
}

# Internal Validation for FLDA
result_flda_internal <- flda(df[, 1:4], df$HeartDisease)
flda_internal_error_rate <- result_flda_internal$error_rate

# Internal Validation for Multinomial
mn_model_internal <- multinom(HeartDisease ~ ., data = df, trace = FALSE)
pred_mn_internal <- predict(mn_model_internal, df)
t_mn_internal <- table(df$HeartDisease, pred_mn_internal)
print(t_mn_internal)
mn_internal_error_rate <- mean(pred_mn_internal != df$HeartDisease) * 100

internal_comparison <- data.frame(
  Method = c("FLDA", "Multinomial"),
  Internal_Error = c(flda_internal_error_rate, mn_internal_error_rate)
)
print(internal_comparison)

# Leave-One-Out Validation (External Validation)
# Suppress iteration output
ctrl <- trainControl(method = "LOOCV", verboseIter = FALSE)

# Leave-One-Out Validation for FLDA
flda_loocv <- train(HeartDisease ~ ., data = df, method = "lda", trControl = ctrl)
flda_loocv_error <- 100 * (1 - flda_loocv$results$Accuracy)

# Leave-One-Out Validation for Multinomial
mn_loocv <- train(HeartDisease ~ ., data = df, method = "multinom", trControl = ctrl, trace = FALSE)
mn_loocv_error <- 100 * (1 - mn_loocv$results$Accuracy)

loocv_comparison <- data.frame(
  Method = c("FLDA", "Multinomial"),
  LOOCV_Error = c(flda_loocv_error, mn_loocv_error)
)
print(loocv_comparison)