Association Rule

An Association Rule is a data mining rule that expresses a co-occurrence relationship between two or more itemset patterns in a transaction dataset.

• AKA: Item Association Rule, Market Basket Rule.
• Context:
  • It can (typically) be produced by an Association Rule Learning Task.
  • It can (typically) consist of an antecedent itemset and a consequent itemset connected by an implication operator.
  • It can (typically) be evaluated by an association rule interestingness measure, such as support, confidence, and lift.
  • It can (often) be derived from a frequent itemset pattern discovered during a frequent pattern mining task.
  • It can (often) require a minimum support threshold and a minimum confidence threshold for its generation.
  • It can (often) support market basket analysis tasks for retail transaction datasets.
  • It can range from being a Single-Item Association Rule to being a Multi-Item Association Rule, depending on its itemset cardinality.
  • It can range from being a Positive Association Rule to being a Negative Association Rule, depending on its itemset correlation.
  • It can integrate with a recommendation system for cross-selling tasks.
  • It can inform a decision support system for business intelligence tasks.
  • ...
• Examples:
  • Retail Association Rules, such as:
    • {item contains 'Hotdog'} ⇒ {item contains 'Hotdog Bun'}, for cross-selling tasks.
    • {item contains 'Diaper'} ⇒ {item contains 'Beer'}, for shopping pattern analysis tasks.
  • Web Usage Association Rules, such as:
    • {visited page A} ⇒ {visited page B}, for clickstream analysis tasks.
  • Bioinformatics Association Rules, such as:
    • {gene G1 expressed} ⇒ {gene G2 expressed}, for gene expression analysis tasks.
  • ...
• Counter-Examples:
  • Production Rule, which encodes an expert-defined condition-action mapping rather than a data-derived co-occurrence.
  • Classification Rule, which predicts a class label rather than an itemset co-occurrence.
  • Sequential Pattern, which captures a temporal ordering rather than an unordered co-occurrence.
• See: Frequent Itemset Pattern, Association, Apriori Algorithm, Basket Analysis, Frequent Itemset, Frequent Pattern, Lift, Confidence, Support.

---

References

2011

• (Toivonen, 2011b) ⇒ Hannu Toivonen. (2011). “Association Rule.” In: (Sammut & Webb, 2011) p.49

2007

• (Han et al., 2007) ⇒ Jiawei Han, Hong Cheng, Dong Xin, and Xifeng Yan. (2007). “Frequent Pattern Mining: current status and future directions.” In: Data Mining and Knowledge Discovery, 15(1). doi:10.1007/s10618-006-0059-1
  • The concept of association rule was introduced together with that of frequent pattern (Agrawal et al. 1993).
• (Hamilton, 2007) ⇒ Howard Hamilton. (2007). “[1] Computer Science 831: Knowledge Discovery in Databases, University of Regina.
  • Association Rule: Probability that particular items are purchased together.
    • X ® Y where X Ç Y = 0

---