# Load necessary libraries 
# install.packages("class") 
library(class) 
# install.packages("datasets") 
library(datasets) 


# Example dataset: iris 
data(iris) 

# Make sure data is scaled (standardized) 
iris_scaled <- scale(iris[, -5])  # all features except species 



# Split data into training and test sets 
set.seed(123) 
sample_index <- sample(1:nrow(iris), 0.7 * nrow(iris)) 
train_data <- iris_scaled[sample_index, ] 
test_data <- iris_scaled[-sample_index, ] 
train_labels <- iris$Species[sample_index] 
test_labels <- iris$Species[-sample_index] 

# Run KNN 
k <- 3  # choose k 
predicted_labels <- knn(train = train_data, test = test_data, cl = train_labels, k = k) 




# Confusion Matrix 
table(Predicted = predicted_labels, Actual = test_labels) 

# Accuracy 
mean(predicted_labels == test_labels) 




# Loop over different k values 
k_values <- 1:15 
accuracy <- sapply(k_values, function(k) { 
  pred <- knn(train = train_data, test = test_data, cl = train_labels, k = k) 
  mean(pred == test_labels) 
}) 

# Plot accuracy vs. k 
plot(k_values, accuracy, type = "b", xlab = "k", ylab = "Accuracy") 
best_k <- k_values[which.max(accuracy)] 








# Example: KNN Classification on a Subset of MNIST-like Data 



# Generate synthetic data 
set.seed(123) 
 
# Class 0: random points around (2, 2) 
class0 <- matrix(rnorm(50*2, mean=2, sd=0.5), ncol=2) 
labels0 <- rep(0, 50) 
 
# Class 1: random points around (4, 4) 
class1 <- matrix(rnorm(50*2, mean=4, sd=0.5), ncol=2) 
labels1 <- rep(1, 50) 
 
# Combine the data 
data <- rbind(class0, class1) 
labels <- factor(c(labels0, labels1)) 
 


train_idx <- sample(1:100, 70) 
test_idx <- setdiff(1:100, train_idx) 
 
train_data <- data[train_idx, ] 
train_labels <- labels[train_idx] 
 
test_data <- data[test_idx, ] 
test_labels <- labels[test_idx] 
 


library(class) 
 
k <- 3 
predictions <- knn(train = train_data, test = test_data, cl = train_labels, k = k) 
 


# Plot train data 
plot(train_data, col=as.numeric(train_labels)+1, pch=19, main='Training Data') 
points(test_data, col=as.numeric(predictions)+1, pch=17, cex=1.5) 
legend("topright", legend=levels(labels), col=1:2, pch=19) 
 


table(Predicted=predictions, Actual=test_labels) 
mean(predictions == test_labels)