Efficient Algorithm Based on Itemset-lattice and Bitmap Index for Finding Frequent Itemsets

Abstract

Abstract Finding frequent itemsets is the key problem in association rules mining. A new algorithm based on the lattice theory and bitmap index for mining frequent itemsets is proposed. The algorithm converts the original transaction database to an itemset-lattice in the pre-processing, where each itemset vertex has a label to save its support, and the complicated task of mining frequent itemsets in the database is thus changed to the simpler one that searches vertexes in the lattice. The efficiency of the mining process is enhanced greatly. Since the support counting in the association rules mining incurs a high cost regarding the I/O, a bitmap index technique is used to speed up the counting process. To address the issue that the intact bitmap usually needs a big memory space for storage, the bit vector is partitioned into blocks, which can be encoded as a symbol. This makes the original bitmap more compact in storage and improve the support counting efficiency as well. Finally, experimental and analytical results are presented.