Association rule learning is a rule-based machine learning method for discovering interesting relations between variables in large databases. It is intended to identify strong rules discovered in databases using some measures of interestingness.

• https://en.wikipedia.org/wiki/Association_rule_learning

R -> L or if R then L

Metrics

• Support: frequency which items occur in the database
• Confidence: frequency the items in R are in the transactions that contain L
• Independence (https://en.wikipedia.org/wiki/Independence_(probability_theory))
• Lift (https://en.wikipedia.org/wiki/Lift_(data_mining))
  • = 1 L and R are independent
  • < 1 negative dependence

  • 1 positive dependence

  • When is close to 1, the association rule is normally not useful
• Leverage
  • = 0 L and R are independent
  • < 0 negative dependence

  • 0 positive dependence

• Odds Ratio

Apriori Algorithm library(arules)

• https://en.wikipedia.org/wiki/Apriori_algorithm
  • minsup can be different for each iteration of the Algorithm
  • a lot of the rules that are returned may not be useful
    • we can keep only more general rules (i.e. {A} => {C} rather than {A,B} => {C})

CBA - Classification Based on Association Rules Algorithm

• https://rdrr.io/cran/arulesCBA/man/CBA.html