Association Rule

Definition

Association rules (Agrawal, Imieliński, & Swami, 1993) can be extracted from data sets where each example consists of a set of items. An association rule has the form X → Y, where X and Y are itemsets, and the interpretation is that if set X occurs in an example, then set Y is also likely to occur in the example.

Each association rule is usually associated with two statistics measured from the given data set. The frequency or support of a rule X → Y, denoted fr(X → Y ), is the number (or alternatively the relative frequency) of examples in which X ∪Y occurs. Its confidence, in turn, is the observed conditional probability $P(Y \mid X) = \mbox{ fr}(X \cup Y )/\mbox{ fr}(X)$.

The Apriori algorithm (Agrawal, Mannila, Srikant, Toivonen & Verkamo, 1996) finds all association rules, between any sets X and Y, which exceed user-specified support and confidence thresholds. In association rule mining, unlike in most other learning tasks, the result thus is a set of rules concerning...