禁忌离散粒子群优化的粗糙集属性约简算法  

Rough Set Attribute Reduction Algorithm Based on Tabu Discrete Particle Pwarm Optimization

作　　者：张荣光[1] 胡晓辉[1] 宗永胜 屈应照 

机构地区：[1]兰州交通大学电子与信息工程学院,兰州730070

出　　处：《小型微型计算机系统》2017年第8期1840-1844,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61163009)资助;甘肃省科技支撑计划项目(144NKCA040)资助

摘　　要：针对数据挖掘中大部分数据属性存在冗余且不具备同等重要性,不利于在数据分析中做出简明的决策,对决策表进行属性约简是规则提取和数据挖掘的重要步骤,提出一种改进的的禁忌离散粒子群优化算法.将决策表的条件属性集合作为离散粒子群,同时引入禁忌搜索算法作为局部搜索策略,提高了粒子群的多样性和寻找全局最优解的能力.在不影响分类质量的前提下,通过粒子间的相互作用最小化条件属性集合,从而删除冗余属性简化知识库.最后利用多组数据进行检验,并与其它算法做了对比实验,实验结果表明此算法能够有效的进行属性约简.Since most data attributes in data mining are redundant and do not have equal importance, it is not conductive to make a concise decision in data analysis. However,attribute reduction of the decision table is an important step in the rule extraction and data mining. Therefore, an improved optimization algorithm of tabu discrete particle pwarm optimization. The condition attribute set of decision table is used as the discrete particle swarm,and the tabu search algorithm is introduced as a local search strategy, thus enriching the diversity of particle swarm and enhancing the ability to seek global optimal solution. Without affecting the quality of classification, the condition attribute set is minimized through the interaction of particles, thereby deleting redundant attributes and simplifying the knowledge base. Finally, through examining multiple sets of data, a comparison experiment with other algorithms is made. As is shown,the algorithm is effective for attribute reduction.

关 键 词：决策表 属性约简 离散粒子群 禁忌搜索 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]