Embedded Files
LIFT
LIFT
Lift is defined as a probability that measures whether the items in the rule have any impact on each other ie, or whether is there any relationship between the items.
Lift (A => B) = P(A ∪ B)/P(A)P(B) = Confidence(A => B)/P(B)
If item A has a positive impact on item B then Lift >1 (Positively Correlated)
If item A has a negative impact on item B then Lift <1 (Negatively Correlated)
If item A has no impact on item B then Lift = 1 (Not Correlated)
Eg: Consider this table through which the lift will be illustrated.
Suppose a rule is set to say Eggs => Bacon
In this case the Lift(Eggs => Bacon) = P(Bacons|Eggs)/P(Bacon) = P(Bacons ∪ Eggs)/P(Bacon) P(Eggs) = 0.4/(0.6 * 0.4) = 1.667
If a rule satisfies both the minimum support and minimum confidence criteria, and additionally, if its lift is greater than 1, indicating a positive correlation between the items, then we can consider the rule as strong and compelling. Such rules hold interest as they signify significant associations in the data and can be effectively utilized for recommendation purposes.
Page updated
Google Sites
Report abuse