The Fast Itemset Miner: A detailed analysis of the candidate generation stages

Association Rule Mining (ARM) represents an important tool for Data Mining techniques. A key step in ARM is to determine the frequent itemsets present within the analyzed data. Recent algorithms addressing this problem focus on identifying frequent itemsets without candidate generation steps. This paper details a new algorithm - Fast Itemset Miner (FIM for short) - that still relies on candidate generation stages, but has a different approach from the standard Apriori algorithm. This approach focuses on increasing size of the interval that candidates belong to, rather than increasing the size of the candidates by one with each corresponding iteration. By comparing the Apriori and FIM candidate generation stages, we show that this second approach uses a faster and more efficient method of determining candidate itemsets. Moreover, such an approach also favors a better support counting method, which greatly impacts on the overall time response of the FIM algorithm.