Hence the tea ∧ latte ⇒ muffin has a support of 0.0625, a confidence of 0.75, and a
lift of 4.5.
If we also assume that Ntea∧muffin = 25, then we can deduce that the rule tea ⇒
muffin has a support of 0.104167, a confidence of 0.5, and a lift of 3. Hence, this
more compact rule has a better support but lower confidence and lift.
Let us also assume that Nlatte∧muffin = 35. This implies that the rule tea ⇒ latte ∧
muffin has a support of 0.0625, a confidence of 0.3, and a lift of 2.057. This rule has a rather poor performance compared to the original rule tea ∧ latte ⇒ muffin: the
support is the same, but the confidence and lift are much lower.
To systematically generate association rules, one typically defines two parameters: minsup and minconf . The support of any rule X ⇒ Y should be above the
threshold minsup, i.e., support(X ⇒ Y) ≥ minsup. Similarly, the confidence of any rule X ⇒ Y should be above the threshold minconf , i.e., confidence(X ⇒ Y) ≥
minconf . Association rules can now be generated as follows: