An Efficient Approach For Candidate Set Generation

When Apriori was first introduced as an algorithm for discovering association rules in a database of market basket data, the problem of generating the candidate set of the large set was a bottleneck in Apriori's performance, both in space and computational requirements. At first, many unsuccessful attempts were made to improve the generation of a candidate set. Later, other algorithms that out performed Apriori were developed that generate association rules without using a candidate set. They used the counting property of association rules instead of generating the candidate set as Apriori does. However, the Apriori concept has been used in many different areas other than counting market basket items, and the candidate generation problem remains a bottleneck issue. The approach described here improves the overall time and space requirements by eliminating the need for a hash table/tree of formation for the candidate set.