Clustering

# Prepare data (scale features)
data <- scale(iris[, 1:4])

# Run K-Means with 3 clusters
set.seed(42)
km <- kmeans(data, centers = 3, nstart = 25)

# Results
km$cluster     # Cluster assignments
km$centers     # Cluster centroids
km$tot.withinss  # Total within-cluster sum of squares

# Elbow method to find optimal K
library(factoextra)
fviz_nbclust(as.data.frame(data), kmeans, method = "wss")

# Visualize clusters
fviz_cluster(km, data = data)

# Compute distance matrix
dist_mat <- dist(data, method = "euclidean")

# Hierarchical clustering
hc <- hclust(dist_mat, method = "ward.D2")

# Plot dendrogram
plot(hc, cex = 0.6)
rect.hclust(hc, k = 3)  # Draw boxes around 3 clusters

# Cut tree to get cluster assignments
clusters <- cutree(hc, k = 3)

# Better dendrogram with factoextra
fviz_dend(hc, k = 3, rect = TRUE, cex = 0.5)

library(dbscan)

# DBSCAN - density-based clustering
db <- dbscan(data, eps = 0.8, minPts = 5)
db$cluster  # 0 = noise points

# Find optimal eps with k-NN distance plot
kNNdistplot(data, k = 5)
abline(h = 0.8, col = "red", lty = 2)

# Visualize
fviz_cluster(db, data = data, geom = "point")

# PCA
pca <- prcomp(iris[, 1:4], scale. = TRUE)

# Variance explained
summary(pca)
fviz_eig(pca)  # Scree plot

# Biplot
fviz_pca_ind(pca, col.ind = iris$Species, addEllipses = TRUE)
fviz_pca_biplot(pca, col.ind = iris$Species)

library(Rtsne)

# Remove duplicate rows first
unique_data <- unique(iris[, 1:4])
set.seed(42)
tsne <- Rtsne(as.matrix(unique_data), dims = 2, perplexity = 30)

# Plot
tsne_df <- data.frame(tsne$Y, Species = iris$Species[!duplicated(iris[,1:4])])
ggplot(tsne_df, aes(X1, X2, color = Species)) +
  geom_point(size = 2) +
  theme_minimal() +
  labs(title = "t-SNE Visualization")

# Silhouette plot for K-Means
fviz_silhouette(silhouette(km$cluster, dist_mat))

# Optimal clusters using silhouette
fviz_nbclust(as.data.frame(data), kmeans, method = "silhouette")