基于类标签变化的改进SLIQ算法研究  

Research on Improved SLIQ Algorithm Based on Changes of Class Label

作　　者：朱王晓嘉 余建坤[1] 

机构地区：[1]云南财经大学,信息学院

出　　处：《微型电脑应用》2015年第10期27-31,4-5,共5页Microcomputer Applications

基　　金：云南省高校商务智能科技创新团队(42212217010)

摘　　要：针对数据挖掘的决策树分类技术中,SLIQ分类器在建树阶段寻找最佳分裂属性时,需要计算大量数值型属性间中间值的基尼系数,算法时间效率低的问题,提出一种改进的SLIQ算法。该算法通过判断数值型属性的预排序属性表中的类标签变化来选择合适分裂位置,减少可能存在的最佳分裂点。实验部分中,用UCI机器学习库中的数据集作分类测试。与原来的SLIQ算法相比,在没有降低分类准确率与扩大决策树规模的情况下,需要计算基尼系数的分裂点个数平均减少了36.32%。最后,将改进算法应用于电子商务的客户分析,分类结果有助于商家作出正确决策。In the technology of decision tree classification in data mining,that large amounts of median Gini Indexes among numeric attributes need to be calculated results in low efficiency of algorithm time when the SLIQ classification searches for the optimal splitting attribute at the stage of tree establishment.Aiming at this problem,an improved SLIQ algorithm is proposed.This algorithm selects appropriate splitting position by judging the changes of class label from the pro-sorting table of numeric attributes to reduce possible existing optimal splitting points.In the experiment,it uses the data sets form UCI machine learning library to do the classification tests.Compared with the traditional SLIQ algorithm,the number of splitting points whose Gini Indexes need to be calculated reduces by 36.32 percent without lowing classification accuracy or expanding the decision tree.At last,after applying the improved algorithm into customer analysis of electronic commerce,the results of classification help the merchants making correct decision.

关 键 词：数据挖掘 决策树 分类 SLIQ算法 分裂点 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]