EBPA: An efficient data structure for frequent closed itemset mining

Abstract

In closed itemset mining, the process of mining from a large transaction database directly often leads to inefficient space and time. Practically, many data structures were proposed to maintain valuable data for frequent closed itemset mining (FCIM), while each data structure has its own advantages and disadvantages. In recent study, a collaboration of array, bitmap, and prefix tree was proposed to gain advantages of those basic data structures by reducing the computing time of the FCIM. That collaboration can save space over that of the original prefix tree, which requires extra space for (m-1) parent-child pointers and its corresponding hashing table (in each tree-node). However, the extra sorting all transactions and merging repeated transactions are required before constructing the prefix tree. Therefore, this paper presents the improved collaboration data structure, called the EBPA (Efficient Bitmap-Prefix-tree Array), with the efficient (parent-child) access in O(1) by using a (temporary) pointer array (without space for hashing) and not require the extra sorting transactions. In system performance evaluation, experimental results showed that the response time of our EBPA-based FCIM mining outperforms over that of the existing collaboration-based FCIM approach.