A conceptual model for multi-level mining and visualization of association rules

Association Rule Mining has been widely used by researchers and practitioners for discovering meaningful rules from large datasets. Recently, there has been an increasing interest in applying association rule mining on data warehouses to identify trends and patterns that exist in the historical data present in large warehouses. However, the application of rule mining algorithms on data warehouses is not a straight forward task. The reason is that the underlying data in a warehouse is modeled in the form of a multidimensional schema, usually the STAR schema, which imposes difficulties in mining rules from its multidimensional structure. Moreover, the data aggregates are stored in the form of data cubes and the presence of dimensional hierarchies makes it even harder to apply rule mining at multiple levels of data abstraction. In this paper, we review the techniques proposed in the literature for mining association rules from data warehouses. Moreover, we critically evaluate the work done in this area and highlight the major limitations and research gaps present in the literature. Literature review reveals that majority of the prior approaches heavily rely on domain knowledge, lack automatic discovery methods, incapable of mining rules at multiple levels of data abstraction, deficient in applying advanced rule interestingness measures and do not provide any visual assistance to analysts for the exploration of discovered rules. In order to overcome these limitations and to fill the identified research gaps, we propose a conceptual model for the discovery of multi-level mining and visualization of association rules from data warehouses. However, the implementation of our proposed model is beyond the scope of this paper.