一种挖掘数值属性的二维优化关联规则方法

摘要

Optimized association rules are permitted to contain uninstantiated attributes.The optimization procedure is to determine the instantiations such that some measures of the roles are maximized.This paper tries to maximize interest to find more interesting rules.On the other hand,the approach permits the optimized association rule to contain uninstantiated numeric attributes in both the antecedence and the consequence.A naive algorithm of finding such optimized rules can be got by a straightforward extension of the algorithm for only one numeric attribute.Unfortunately,that results in a poor performance.A heuristic algorithm that finds the approximate optimal rules is proposed to improve the performance.The experiments with the synthetic data sets show the advantages of interest over confidence on finding interesting rules with two attributes.The experiments with real data set show the approximate linear scalability and good accuracy of the algorithm.%优化关联规则允许在规则中包含未初始化的属性.优化过程就是确定对这些属性进行初始化,使得某些度量最大化.最大化兴趣度因子用来发现更加有趣的规则;另一方面,允许优化规则在前提和结果中各包含一个未初始化的数值属性.对那些处理一个数值属性的算法进行直接的扩展,可以得到一个发现这种优化规则的简单算法.然而这种方法的性能很差,因此,为了改善性能,提出一种启发式方法,它发现的是近似最优的规则.在人造数据集上的实验结果表明,当优化规则包含两个数值属性时,优化兴趣度因子得到的规则比优化可信度得到的规则更有趣.在真实数据集上的实验结果表明,该算法具有近似线性的可扩展性和较好的精度.