Interestingness Measure: Correlation Lift

* Measure of dependent/correlated events: lift

 

1) Statistiacal independence

 

P(S∩B) = P(S) x P(B) => Statistically independence

P(S∩B) > P(S) x P(B) => Postively correlated

P(S∩B) < P(S) x P(B) => Negatively correlated

 

2) lift : interestingness measure: (interesting = reasonable 정도로 이해하면 될 것 같다)

 

 

 

- if the lift is equal to 1, then A and B are independent and there is no correlation between them.

   A⊥B = [lift(A,B) = 1] = [cov(A,B) = 0]

 

- lift (A,B) = lift(B,A),

 

ex)

 

 

 

- lift (B,C) = (2000/5000)/[(3000/5000)*(1250/5000)] = 0.89 < 1  : negatively correlated

 

- lift (B,C_bar) = (1000/5000)/[(3000/5000)*(1250/5000)] = 1.33 > : positively correlated