Optimising Association Rule Algorithms Using Itemset Ordering

Association-rule mining is a well-known method of Knowledge Discovery in Databases, aimed at identifying observable relationships in datasets the records of which can be represented as sets of items The problem of extracting all possible association rules, however, is computationally intractable, so methods of reducing its complexity are of continuing interest. We here describe some results obtained from a method we have developed, which reduces the task by means of an efficient restructuring of the data accompanied by a partial computation of the totals required. The method is sensitive to the ordering of items in the data, ai d our experiments show how this property can be used as an implement.etion heuristic. The results we show demonstrate the performance gain for our method in comparison with a benchmark alternative, and the further gain from using the ordering heuristic