基于SPRINT方法的并行决策树分类研究

Study on the parallelism of decision tree classification based on SPRINT

决策树技术的最大问题之一就是它的计算复杂性和训练数据的规模成正比,导致在大的数据集上构造决策树的计算时间太长。并行构造决策树是解决这个问题的一种有效方法。文中基于同步构造决策树的思想,对SPRINT方法的并行性做了详细分析和研究,并提出了进一步研究的方向。

One of the greatest problems with the decision tree technique is that the computational complexities are normally proportional to the size of training data set, so the computing time of constructing decision tree on large data sets is prohibitive. Parallelism is one effective solution to this problem. In this paper, parallel formulation of SPRINT decision tree algorithm based on synchronous tree construction approach was presented and a proposal of further study was given.