Real World Association Rule Mining

Across a wide variety of fields, data are being collected and accumulated at a dramatic pace, and therefore a new generation of techniques has been proposed to assist humans in extracting usefull information from the rapidly growing volumes of data. One of these techniques is the association rule discovery, a key data mining task which has attracted tremendous interest among data mining researchers. Due to its vast applicability, many algorithms have been developed to perform the association rule mining task. However, an immediate problem facing researchers is which of these algorithms is likely to make a good match with the database to be used in the mining operation. In this paper we consider this problem, dealing with both algorithmic and data aspects of association rule mining by performing a systematic experimental evaluation of different algorithms on different databases. We observed that each algorithm has different strengths and weaknesses depending on data characteristics. This careful analysis enables us to develop an algorithm which achieves better performance than previously proposed algorithms, specially on databases obtained from actual applications.