Efficient quantitative frequent pattern mining using predicate trees

Databases information is not limited to categorical attributes, but also contains much quantitative data. The typical approach of quantitative association rule mining is achieved by using intervals. Such approach involves many interval generations and merges, and requires reconstruction of tree structures, such as hash trees, R-trees, and FP-trees. When intervals change, these tree structures are re-constructed on-the-fly, which is very time-consuming. In this paper, we present a Predicate tree based quantitative frequent pattern mining algorithm (PQM). The central idea of PQM is to exploit Predicate trees to facilitate quantitative frequent mining without tree re-construction. Our method has three major advantages: 1) P-trees are pre-generated tree structures, which are flexible and efficient for any data partition and any interval optimization; 2) PQM is efficient by using fast P-tree logic operations; and 3) PQM has better performance due to the vertically decomposed structure and compression of P-trees. Experiments show that our algorithm outperforms Apriori algorithm by orders of magnitude with better support threshold scalability and cardinality scalability.