数量关联规则发现中的聚类方法研究

应用聚类方法研究了数量关联规则提取过程中的连续属性离散化问题，由于现存的方法倾向于将支持度较高的区域划分为多个区间，对高偏数据效果不理想，针对这一问题，提出聚类算法ＰＫＣＣＡ，与传统快速聚类不同，ＰＫＣＣＡ在迭代过程中动态调整中心个数，避免造成小支持度问题，并继承了传统快速聚类适合大样本的优点。

Clustering Method for Mining Quantitative Association Rules

This paper presents a cluster method for discretization in the processing of mining quantitative association rules. Many efficient algorithms for mining association rules were presented in recent years. But association rules containing quantitative attributes is still a bottle neck in these works. Agrawal presented partial k completeness method for partitioning. Fukuda presented the equi depth partitioning and algorithm based on gaining optimized regions. Current method doesn't work very well on highly skewed data because it tends to split intervals with relative high support into several intervals, even if their behaviors are very similar. In order to solve this problem, this paper presents a Cluster Algorithm named PKCCA. It is different from traditional Cluster Algorithm on that PKCCA dynamically adjust the number of median in iterations, but it still fit for large sample that was the character of fast clustering. For efficiency consideration, sampling method is adopted in this paper.