Fast Discovery of Frequent Itemsets: a Cubic Structure-Based Approach

Mining frequent patterns in large transactional databases is a highly researched area in the field of data mining. The different existing frequent pattern discovering algorithms suffer from various problems regarding the computational and I/O cost, and memory requirements when mining large amount of data. In this paper a novel approach is introduced for solving the aforementioned issues. The contribution of the new method is to count the short patterns in a very fast way, using a specific index structure. The suggested algorithm is partially based on the apriori hypothesis and exploits the benefit of a new index table-based cubic structure to count the occurrences of the candidates. Experimental results show the advantageous execution time behavior of the proposed algorithm, especially when mining datasets having huge number of short patterns. Its memory requirement, which is independent from the number of processed transactions, is another benefit of the new method.