Discover, recycle and reuse frequent patterns in association rule mining

Abstract

In this thesis, an extended framework of mining and recycling frequent patterns for association rule mining is given. Within the framework, several technical problems are addressed. First, a novel frequent pattern algorithm is proposed to mine microarray datasets with large number of columns and small number of rows. The new algorithm is orders of magnitude better than existing algorithms. Second, the concept of interesting rule groups is proposed to reduce the enormous number of redundant rules discovered from microarray data and an efficient algorithm is proposed to mine interesting rule groups. The new algorithm outperforms the existing rule mining algorithms by several orders of magnitudes. Third, two approaches of recycling previous frequent patterns are proposed to speed up the subsequent rounds of mining in the constrained association rule mining, which is an iterative process. The first method recycles the previous frequent patterns and intermediate results to speed up subsequent mining process when constraints are changed while the second method only needs previous frequent patterns for recycling.