A New Approach for the Discovery of Frequent Itemsets

The discovery of the most recurrent association rules, in a large database of sales transactions requires that the sets of items bought together by a sufficiently large population of customers are identified. This is a critical task, since the number of generated itemsets grows exponentially with the total number of items. Most of the algorithms start identifying the sets with the lowest cardinality, and subsequently, increase it progressively. Our approach is different, since the sets to be considered at a time are determined by the items in the sets. The main advantage is a significant reduction of the CPU time required to update data structures in main memory. This paper presents an algorithm that requires only one pass on the database, presents linear scale-up property with the dimensions of the database and, as shown by the experiments, performs better than other classical algorithms.