一种多属性约简支持向量机混合分类方法.PDF

31 4 ( ) V o.l 31 N o4 2006 8 Journal ofKunm ingU niversity of Science and Tec nology ( Science and Tec nology) A ug. 2006 李 波, 郭凤菊, 李新军 ( , 300072) : 针对属性特别多仅用一种属性约简方法难以实现有效约简的情况, 提出了基于双重属性 约简的 合支持向量机分类方法. 通过引入贡献率和正确率两 个概念, 首先采用主成分分析算法 计算各 个条件属性的贡献率, 根据贡献率大小和给定的阈值去掉条件属性中贡献率小的成分, 提 取信息量最大的主要成分; 然后再基于粗糙集的属性约简理论, 计算这些主要成分对决策变量的 正确率, 对这些属性进行第二次约简; 该方法采用定性定量相结合的方式, 可以最大程度地去除 属性集中冗余的或不重要的属性, 保证将最简的属性样本集输入支持向量机进行建模预测. 最后 的仿真试验验证了我们所提方法的有效性和正确性. : 粗糙集; 支持向量机; 合分类算法 : TP18 : A : 1007- 855X ( 2006) 04- 0113- 05 A Hybrid SupportVectorM achine ClassificationM ethod Based on DualAttribute Reductions L I Bo, GUO Feng-ju, L IX in-jun ( Sc ool ofManagement, T ianjin Un iversity, T ianjin 300072, C ina) Abstract: A ybrid support vectormac ine classification algorit m based on dual attribute reductions is proposed on t e condition t at it is difficult to ac ieve effective reduction to numerous attributes only by one algorit m of attribute reduction. In t is algorit m, wt o defin itions about t e selection of attribute mi portance- correlation de- gree and contribution degree are introduced. F irstly, correlation degree of eac condition attribute is computed based on principal component analysis ( PCA ) and t en t e most informative principal components are selected t roug deleting secondary components based on t e correlation degree and t e given t res old. Secondly, t e contribution degree of selected attributes in t e process of PCA to decision variances is computed and t en t e at- tribute reduction is mi plemented again on t e basis of roug sets. In t e ybrid algorit m, t e comb ination of qualitative and quantitative is ap