基于vprs多变量决策树优化算法-计算机系统应用.pdf

算 机 系 统 应 用 2010 年 第 19卷 第 12 期 基于VPRS 多变量决策树优化算法① 邱云飞 王 光 关晓林 邵良杉 (辽宁工程技术大学系统工程研究所 辽宁 葫芦岛 125105) 要： 噪声数据降低了多变量决策树的生成效率和模型质量，目前主要采用针对叶节点的剪枝策略来消除 噪声数据的影 ，而对决策树生成过程中的噪声干扰问题却没有给予关注。为改变这种状况，将基本 粗糙集(rough set，RS)理论中相对核的概念推广到变精度粗糙集(variable precision rough set， VPRS)理论中，并利用其进行决策树初始变量选择；将两个等价关系相对泛化的概念推广为两个等价 关系多数包含情况下的相对泛化，并利用其进行决策树初始属性检验；进而给出一种能够有效 除噪 声数据干扰的多变量决策树构造算法。最后，采用实例验证了算法的有效性。 关键词：单变量决策树；多变量决策树；噪声数据；变精度粗糙集；相对核 Optimization Algorithm for Multivariate Decision Trees Based on VPRS QIU Yun-Fei, WANG Guang, GUAN Xiao-Lin, SHAO Liang-Shan (Institute of System Engineering, Liaoning Technical University, Huludao 125105, China) Abstract: When construct multivariate decision trees, noise data reduced the training efficiency and quality of model, most of the present pruning methods aimed at leaf node to eliminate the influence of noise data, but not pay attention to the disturbed problem of noise data when selected testing attribute. In order to solve the problem, extends the relative core of attributes in rough sets theory to variable precision rough set(VPRS), and uses it for selection of initial variables for decision tree; extends the concept of generalization of one equivalence relation with respect to another one, to relative generalization equivalence relation under mostly-contained condition, and uses it for decision tree initial attribute check;propose an algorithm for multivariate decision tree that can avoid disturbance of noisy data. Finally, validated the algorithm by an experiment. Keywords: univariate decision trees; multivariate decision trees; noisy data; variable precision rough set; relative core of attributes 决策树学习是以实