基于基因表达式编程属性融合分类算法.pdfVIP

基于 因表达式编程的属性融合分类算法1 赵波 唐常杰 朱明放 魏大刚 ( 四川大学计算机学院 成都 610065) ( {Zhaobo,tangchangj ie}@) 摘要 传统符号分类算法 （如C4.5 ）在处理属性间有较大关联的数据集时不容易获得好的分 类效果。为了改进传统符号分类算法的不足，本文做了如下探索：（1）提出了属性融合概念 和具有属性融合特色的进化适应度函数，在此基础上设计了基于基因表达式编程的属性融合 分类算法(GEP Attribute Merge, GAM) ；（2 ）提出了保留精英基因的策略（Saving Best_Gene ）， 加快GAM 算法的收敛速度；（3 ））提出了基于“折半”原理的准小属性集寻找算法(Subsmall Attribute Set,SAS)；（4 ）通过比较实验验证了GAM 算法的有效性，使得分类精度较传统符 号分类算法平均提高了约 30%,保留精英基因策略可以加快GAM 算法的收敛速度，在同进 化代数条件下平均最优适应度提高了约3%, 以及SAS 算法可以发现准小属性集。 关键词 基因表达式编程；属性融合；精英基因；准小属性集 Attribute-Merge Classification Based on Gene Expression Programming ZHAO Bo , TANG Changji e, ZHU Mngfang WEI dagang (School of Computer Science, Sichuan University, Chengdu 610065) Abstract Traditional symbolic classification algorithms, for instance C4.5, can not deal with the relationships among attributes in dataset. For solving the problem, the contributions of this paper include: ( 1) Proposing a new concept called Attribute Merge, Fitness Function with Attribute Merge and a new algorithm called GEP_based Attribute Merge(GAM) on this new concept; (2) Designing a new method called Saving Best_Gene(SBG) for better convergence; (3) Proposing a new approach called Subsmall Attribute Set(SAS) for finding a sub optimal attribute set; (4) Giving extensive experiments to show that GAM is effective, on the contrary to traditional symbolic classification algorithms the average accuracy can be improved by 30%, and convergence of GAM can be improved by saving the best gene, under the same generation, the average best_fitness can be improved by 3%, and then sub-small attribute set can be discovered by SAS. Keywords: GEP, Attribute Merge, Best Gene, Sub-small Attribute Set 1. 引 言 分类是数据挖掘中典型的知识提取算法。现实世界中一个对象的多个属性往往相关，使 基于属性不相关假定的符号分类算法束手无策。符号分类算