- Association rule mining 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. [Wikipedia]
- e.g. If item A is bought by customer c1 then chances of being bought item B by same customer on same transaction.
- A == > B, if A then B; A = antecedent and B = consequent.
- If we have huge dataset(transaction) then we can get so many combination of type A => B , A&B => C, A&B&C => D and so on so to find promising rules we need some metrices. Those metrices are follows:
- Support :- support is frequency of item or items combination.
- supp(X)=Number of transaction in which X appears / Total number of transactions
- Confidence: - How often items collection B occur given occurrence of items collection A.
- conf(X⟶Y)=supp(X∪Y) / supp(X)
- Lift:- it’s Strength of any rule.
- lift(X⟶Y)=supp(X∪Y) / supp(X)∗supp(Y)
Where Map Reduce is Used?
- It is used for finding frequencies (to find value of support) of X in transactions.
References: