Performing cluster analyis in R

clusteR.R

# Cluster analysis in R
# inspired by Dima Gorenshteyn, DataCamp

## standardize data
df_st <- scale(df)

## Hierachical clustering
d <- dist(df)
hc <- hclust(d, "method") # method %in% c("complete", "average", "single")
c <- cutree(hc, h = the_height) # h: the height to cut the tree # assign cluster

s_df <- df %>% mutate(c) # segment data frame
s_df %>% count(c)

dend_g <- as.dendrogram(hc) # create dendrogram from hc
dend_g_c <- color_branches(dend_g, h = the_height) # the color of the dendrogram can be configured by the height

plot(dend_g_c)

## K-means clustering
k_m <- kmeans(df, centers = k) # define k
k_df <- df %>% mutate(k_m$cluster)

### K-means: Elbow method
tot_withinss <- purrr::map_dbl(1:(n_var-1),  function(k){ # max(k) = variable (n_var - 1)
  model <- kmeans(x = lineup, centers = k)
  model$tot.withinss
}) # total within-cluster sum of squares

elbow_df <- data.frame(k = 1:(n_var-1), tot_withiness)

ggplot(elbow_df, aes(x = k, y = tot_withinss)) +
  geom_line() +
  scale_x_continuous(breaks = 1:10)

### K-means: Silhouette [silwɛt] analysis 
# within cluster distance
# 
m_pan <- cluster::pam(df, k)
cluster::silhouette(m_pan) %>% plot()