An Algorithm Based On Density And Grid For Mining And Clustering Association Rules

It is one of the important patterns in data mining research to mine association rules, which is used to find out the dependence between different attributes for given support and confidence threshold values. Many classical and improved algorithms have been proposed for mining Boolean association rules. But the number of the rules mined by all these algorithms is not too large owing to the inherent "0-1"property of Boolean attribute values. When quantitative association rules mined, the tremendous problem of searching time and space is certain to be faced with, due to the succession of quantitative attribute values. A great deal of work has been done to partition quantitative attribute, and improved mining algorithms performance in time and space. However the problem of a large number of association rules returned by algorithm has not been solved well. In this paper, on the base of classical clustering algorithm, combining the advantage that density-based method can find out clusters in random shape and grid-based method has the capacity of rapid execution, a clustering algorithm based on grid and density is proposed to perform mining and clustering quantitative association rules. By combining similar, adjacent association rules to form a few general rules, the clustered association rules found are helpful in reducing the large number of association rules that are typically computed by existing algorithm, thereby tendering the clustered rules much easier to interpret and visualize.The algorithm begins by taking source data in tuple form and partitioning those attributes that take values from a continuous domain. Then the algorithm performs a single pass over the data using an association rule engine to derive a set of association rules. Next, all those two-attribute association rules where the right-hand side of the rules satisfies our segmentation criteria are clustered. The segmentation for accuracy is tested, and if necessary certain system parameters are modified to produce a better segmentation and repeat the process.To verify the validity and correctness of the algorithm, an association rules clustering system is constructed, in which we apply the new method besides giving the relevant theory. According to the theoretical analysis and the result of experiment, the algorithm apparently speeds the execution of mining quantitative association rules, reduces the number of association rules we finally find, making the rules easier to understand.