1. Create I1. This is the set of singleton frequent item-sets, i.e., item-sets with a support above the threshold minsup containing just one element.
2. k := 1.
If Ik = ∅, then output /k
Ii and end. If Ik /= ∅, continue with the next
step.
Create Ck+1 from Ik . Ck+1 is the candidate set containing item-sets of cardinality k + 1. Note that one only needs to consider elements that are the union of two item-sets A and B in Ik such that |A ∩ B|= k and |A ∪
B|= k + 1.
For each candidate frequent item-set c ∈ Ck+1: examine all subsets of c
with k elements; delete c from Ck+1 if any of the subsets is not a member
of Ik .
For each item-set c in the pruned candidate frequent item-set Ck+1, check
whether c is indeed frequent. If so, add c to Ik+1. Otherwise, discard c.
k := k + 1 and return to Step 3.
The algorithm only considers candidates for Ik+1 that are not ruled out by evidence in Ik . This way the number of traversals through the data set is reduced dramatically.