一种基于纵向划分数据集的并行决策树分类算法

A Parallel Decision Tree Classification Algorithm Based on Vertical Data Partitioning

本文提出了一种处理多属性数据集的快速可扩展性并行分类算法—FSPC算法。它首次采用了纵向划分数据集以及在测试属性的选择过程中同步划分数据集等方法。实验结果表明 ,它不仅有利于减少通信及进行I/O的开销 ,而且有利于提高算法的并行度。

We present a fast scalable parallel classification algorithm in this paper named FSPC to handle large databases with lots of attributes.It is the first algorithm to introduce several kinds of techniques such as partitioning databases vertically,and performing the split while finding split points.Experimental results show that these techniques can not only reduce communication and I/O costs,but also increase the algorithm parallelism.