cluster_associations function

Cluster association rules

This function clusters association rules based on the selected numeric attribute by (e.g., confidence or lift) and summarizes the clusters. The clustering is performed using the k-means algorithm.

Each cluster is represented by a label consisting of the number of rules in the cluster and the most common predicates in the antecedents of those rules.

cluster_associations(
  x,
  n,
  by,
  algorithm = "Hartigan-Wong",
  predicates_in_label = 2
)

Arguments

• x: A nugget of flavour associations, typically the output of dig_associations().

• n: The number of clusters to create. Must be a positive integer.

• by: A tidyselect expression (see tidyselect syntax) specifying the numeric column to use for clustering.

• algorithm: The k-means algorithm to use. One of "Hartigan-Wong"

  (the default), "Lloyd", "Forgy", or "MacQueen". See stats::kmeans()

  for details.

• predicates_in_label: The number of most common predicates to include in the cluster label. The default is 2.

Returns

A tibble with one row per (cluster, consequent) pair. The columns are:

• cluster: the cluster number;
• cluster_label: a label for the cluster, consisting of the number of rules in the cluster and the most common predicates in the antecedents of those rules;
• consequent: consequents of the rules;
• other numeric columns from the input nugget, aggregated by mean within each cluster.

Examples

# Prepare the data
cars <- mtcars |>
    partition(cyl, vs:gear, .method = "dummy") |>
    partition(carb, .method = "crisp", .breaks = c(0, 3, 10)) |>
    partition(mpg, disp:qsec, .method = "triangle", .breaks = 3)

# Search for associations
rules <- dig_associations(cars,
                          antecedent = everything(),
                          consequent = everything(),
                          max_length = 3,
                          min_support = 0.2)

# Cluster the found rules
clu <- cluster_associations(rules, 10, "lift")

# Print the number of clusters
length(unique(clu$cluster))

## Not run:

# Plot the clustered rules
library(ggplot2)

ggplot(clu) +
   aes(x = cluster_label, y = consequent, color = lift, size = support) +
   geom_point() +
   xlab("predicates in antecedent groups") +
   scale_y_discrete(limits = rev) +
   theme(axis.text.x = element_text(angle = 45, hjust = 1))
## End(Not run)

See Also

dig_associations(), association_matrix() stats::kmeans()

Author(s)

Michal Burda