Framework for mining of association rules from data warehouse

association rules, data warehouse, data mining

In this paper, we propose a framework for association rules mining from data warehouses. This framework presents alliance between two business intelligence areas. First area is represented by data warehouse and data cube providing high quality data. The second area is represented by data mining, especially association rules mining providing an additional knowledge.
Association rules mining on data warehouses is different from mining on relational or transactional databases, because it deals with couple of dimensions, which form conceptual hierarchies. Thus we mine multi- and inter-dimensional association rules. There are several approaches how to mine such association rules described in literature. This framework presents a novel combination of the data cube processing - top-down (on product dimensions) and bottom-up (on domain dimensions). We presume division of dimensions on domain and product dimensions.
The framework works in the following steps.  The first one represents obtaining frequent leaf 1-itemsets, which means obtaining frequent itemsets from domains represented by items from domain dimensions on leaf level. In the second step we obtain all frequent 1-itemset. Following step represents iterative mining of frequent k-itemset from frequent (k-1)-itemsets. In the final step we process all k-itemsets and obtain association rules from them.