Lift is arguably the most important of the 3 measures; it measures the support of the rule relative to the support of the individual sides of the expression; put differently, it measures how strong the rule is with respect to a random occurrence of the LHS and RHS of the expression. It is formally defined as:

Lift = Support (Rule)/(Support(LHS) * Support (RHS))

A low value for lift (say, less than or equal to 1) indicates that the LHS and RHS occurrence are independent of one another, whereas a higher lift measure indicates that the co-occurrence is significant.

In our prior example,

{Bread} --> {Butter} has a lift of:

Support ({Bread} --> {Butter})
Support {Bread} * Support {Butter}

= 0.50/((3/4) * (3/4)) = 0.50/(0.75 * 0.75) = 0.89.

This indicates that although the Confidence of the rule was high, the rule in and of itself is not significant relative to other rules that may be higher than 1.

An example of a rule with a Lift higher than 1 would be:

{Item 1: Bread} --> {Item 3: Cheese}

This has a Lift of:

Support {Item 1: Bread --> Item 3: Cheese}/(Support {Item 1: Cheese} * Support {Item 3: Cheese})

= (1/4)/((1/4)*(1/4) = 4.