基于MapReduce 的ACO-K-means 并行聚类算法.pdfVIP

1 Computer Engineering and Applications 计算机工程与应用 2013 ，49 （16） 117 基于MapReduce 的ACO-K-means 并行聚类算法 虞倩倩，戴月明，李晶晶 YU Qianqian, DAI Yueming, LI Jingjing 江南大学 物联网工程学院，江苏 无锡 214122 School of IOT, Jiangnan University, Wuxi, Jiangsu 214122, China YU Qianqian, DAI Yueming, LI Jingjing. Parallel ACO-K-means clustering algorithm based on MapReduce. Computer Engineering and Applications, 2013, 49 （16）：117-120. Abstract ：There is a serious lack of memory when use K -means to deal with massive data. In this paper MapReduce is used to parallelize K -means. Due to ordinary K -means has slow rate of convergence, easily fall into local optimization and the limita- tions of selection of initial cluster centers, Ant Colony Optimization （ACO ）is led into K -means. The result demonstrates ACO-K -means clustering algorithm based on MapReduce model has high speedup and good scalability, and it ’s convergence and clustering accuracy are also improved in some degree. Key words ：data mining; MapReduce; Ant Colony Optimization （ACO ）; K -means; cloud computing 摘 要：针对K -means 算法处理海量数据存在严重的内存不足，提出利用MapReduce 并行化K -means ，但是普通的K 均值 存在收敛速度慢、易陷入局部最优和对初始聚类中心的选取等局限性，因此选择了经ACO 改进过的ACO-K -means 聚类算 法。实验结果表明，经MapReduce 并行化的ACO-K -means ，不仅具有良好的加速比和扩展性，其收敛性以及聚类精度均得 到了改善。 关键词：数据挖掘；MapReduce ；蚁群优化；K -means；云计算 文献标志码：A 中图分类号：TP301 doi ：10.3778/j.issn. 1002-8331.1204-0243 - C (j 12 k) 1 引言 是待聚类的数据集合，j 为聚类中心，初始 所谓数据挖掘（Data Mining ），就是从海量数据中获取 值任意分配；