A time- and memory-efficient frequent itemset discovering algorithm for association rule mining

Frequent itemset discovering is a highly researched area in the field of data mining. The algorithms dealing with this problem have several advantages and disadvantages regarding their time complexity, I/O cost and memory requirement. There are algorithms that have moderate memory usage but high I/O cost, thus the execution time of them is high; such methods are for example the level-wise algorithms. Other methods have advantageous time behaviour; however, they are memory intensive, like the two-phase algorithms. In this paper, a novel algorithm, which is efficient both in time and memory, is proposed. The new algorithm discovers the small frequent itemsets quickly by taking advantage of the easy indexing opportunity of the suggested candidate storage structure. The main benefit of the novel algorithm is its advantageous time behaviour when using different types of datasets as well as its low I/O activity and moderate memory requirement.