基于病态指数循环分析的评价指标筛选研究
|
摘要
为了避免指标集较高的信息重叠扭曲综合评价结果,已有研究通过剔除相关程度高的少部分(主要是两两)指标中相对不重要的指标,虽可保证指标集内少部分指标间的信息重叠不高,但却无法保证指标集的整体信息重叠不高。为此,本研究提出一种新的评价指标筛选方法。其创新与特色主要有三:一是根据一个指标剔除后指标集病态指数减小的幅度越大,该指标对指标集整体信息重叠贡献越大的思路,剔除对剩余指标整体信息重叠贡献最大的指标,循环往复直至剩余指标的整体信息重叠不高为止,克服已有研究无法高效率地降低指标集整体信息重叠及不能测度剩余指标整体信息重叠水平的不足;二是通过剔除相关程度高的任意两个剩余指标中相对不重要的一个指标,避免指标集整体信息重叠不高而个别指标间信息重叠却依然较高;三是通过指标集病态指数减少比例与指标剔除数量的比值表示指标集整体信息重叠降低的效率,检验信息重叠指标筛选方法的相对合理性,弥补已有研究缺乏相关检验标准的不足。最后,通过一组科技评价指标的筛选说明上述方法的可行性及有效性。
The existing researches delete the relatively unimportant indicators in few indices(mainly including any pair indicators)with high correlation to prevent the high information overlap from distorting the comprehensive evaluation results,which can improve the information overlap of few indicators but is not available to ensure that of the indices set is not high.In this article,a new method of screening indices with high information overlapping is presented.The indicators of the greatest contribution to the overall overlap of residual indicators are deleted successively until the information overlap of all remaining indicators is not high so far,since an indicator makes greater contribution to the information overlap of the indices set if the decrease of ill-condition index for the indices set is lager when it is removed.Furthermore,the dominated one in any pair of remaining indicators with high correlation is deleted to avoid high information overlap between the individual indicators while the overall information overlap of the indices set is not high.The reduction ratio of ill-condition index for the indices set and the number of deleted indicators are used to illustrate the efficiency of reducing the overall information overlap for the indices set,which tests the relative rationality of the suggested method that is more efficient than the correlation analysis method for reducing the overall information overlap for the indices set.The above findings enrich the theoretical research of the indices screening methods,helpfully complements the comprehensive evaluation method,and facilitates the researches in related fields.
The existing researches delete the relatively unimportant indicators in few indices(mainly including any pair indicators)with high correlation to prevent the high information overlap from distorting the comprehensive evaluation results,which can improve the information overlap of few indicators but is not available to ensure that of the indices set is not high.In this article,a new method of screening indices with high information overlapping is presented.The indicators of the greatest contribution to the overall overlap of residual indicators are deleted successively until the information overlap of all remaining indicators is not high so far,since an indicator makes greater contribution to the information overlap of the indices set if the decrease of ill-condition index for the indices set is lager when it is removed.Furthermore,the dominated one in any pair of remaining indicators with high correlation is deleted to avoid high information overlap between the individual indicators while the overall information overlap of the indices set is not high.The reduction ratio of ill-condition index for the indices set and the number of deleted indicators are used to illustrate the efficiency of reducing the overall information overlap for the indices set,which tests the relative rationality of the suggested method that is more efficient than the correlation analysis method for reducing the overall information overlap for the indices set.The above findings enrich the theoretical research of the indices screening methods,helpfully complements the comprehensive evaluation method,and facilitates the researches in related fields.
引文
[1]苏为华.我国多指标综合评价技术与应用研究的回顾与认识[J].统计研究,2012,29(8):98-107.
[2]余肖生,周宁.高维数据降维方法研究[J].情报科学,2007,25(8):1248-1251.
[3]胡永宏.综合评价中指标相关性的处理方法[J].统计研究,2002,3:39-40.
[4]Changki L,Geunbae L G.Information gain and divergence based feature selection for machine learningbased text categorization[J].Information Processing&Management,2006,42(1):155-165.
[5]Zhu Zexuan,Ong Y S,Dash M.Markov blanket-embedded genetic algorithm for gene selection[J].Pattern Recognition,2007,40(11):3236-3248.
[6]Inza I,Larra1aga P,Etxeberria R,et al.Feature subset selection by bayesian network-based optimization[J].Artificial Intelligence,2000,123(1-2):157-184.
[7]Zhang Yudong,Wang Shuihua,Phillipsc P,et al.Binary PSO with mutation operator for feature selection using decision tree applied to spam detection[J].Knowledge-Based Systems,2014,64:22-31.
[8]Aytug H.Feature selection for support vector machines using Generalized Benders Decomposition[J].European Journal of Operational Research,2015,244(1):210-218.
[9]迟国泰,陈洪海.基于信息敏感性的指标筛选与赋权方法研究[J].科研管理,2016,37(1):153-160.
[10]李鸿禧,迟国泰.基于DEA-t检验的以企业为主体的科技创新效率评价[J].中国管理科学,2016,24(11):109-119.
[11]范柏乃.我国城市居民生活质量评价体系的构建与实际测度[J].浙江大学学报(人文社会科学版),2006,36(4):122-130.
[12]张昆,迟国泰.基于相关分析-粗糙集理论的生态评价指标体系构建[J].系统工程学报,2012,27(1):119-128.
[13]迟国泰,曹婷婷,张昆.基于相关-主成分分析的人的全面发展评价指标体系构建[J].系统工程理论与实践,2012,32(1):112-119.
[14]陈洪海,迟国泰.基于信息显著性差异的绿色产业评价指标筛选模型及实证[J].运筹与管理,2014,23(6):193-204.
[15]赵志冲,迟国泰.基于似然比检验的工业小企业债信评级研究[J].中国管理科学,2017,25(1):45-56.
[16]韩伯棠,王莹.中国科技人力资源评价指标体系构建方法研究[J].北京理工大学学报(社会科学版),2006,8(6):31-35.
[17]Destrero A,Mosci S,Mol C D.Feature selection for high dimensional data[J].Computational Management Science,2009,6(1):25-40.
[18]顾雪松,迟国泰,程鹤.基于聚类-因子分析的科技评价指标体系构建[J].科学学研究,2010,28(4):509-514.
[19]周立斌,李刚,迟国泰.基于R聚类-变异系数分析的人的全面发展评价指标体系构建[J].系统工程,2010,28(12):56-63.
[20]赵宇哲,刘芳.生态港口评价指标体系的构建-基于R聚类、变异系数与专家经验的分析[J].科研管理,2015,36(2):124-132.
[21]Chen Honghai,Chi Guotai.Urban green development evaluation indicator system model based on clusteringrough set and application[J].ICIC Express Letters Part B:Applications,2015,6(10):2649-2654.
[22]王斌会.计量经济学模型及R语言应用[M].广州:暨南大学出版社,2015.
[23]郭亚军.综合评价理论、方法及应用[M].北京:科学出版社,2007.
[2]余肖生,周宁.高维数据降维方法研究[J].情报科学,2007,25(8):1248-1251.
[3]胡永宏.综合评价中指标相关性的处理方法[J].统计研究,2002,3:39-40.
[4]Changki L,Geunbae L G.Information gain and divergence based feature selection for machine learningbased text categorization[J].Information Processing&Management,2006,42(1):155-165.
[5]Zhu Zexuan,Ong Y S,Dash M.Markov blanket-embedded genetic algorithm for gene selection[J].Pattern Recognition,2007,40(11):3236-3248.
[6]Inza I,Larra1aga P,Etxeberria R,et al.Feature subset selection by bayesian network-based optimization[J].Artificial Intelligence,2000,123(1-2):157-184.
[7]Zhang Yudong,Wang Shuihua,Phillipsc P,et al.Binary PSO with mutation operator for feature selection using decision tree applied to spam detection[J].Knowledge-Based Systems,2014,64:22-31.
[8]Aytug H.Feature selection for support vector machines using Generalized Benders Decomposition[J].European Journal of Operational Research,2015,244(1):210-218.
[9]迟国泰,陈洪海.基于信息敏感性的指标筛选与赋权方法研究[J].科研管理,2016,37(1):153-160.
[10]李鸿禧,迟国泰.基于DEA-t检验的以企业为主体的科技创新效率评价[J].中国管理科学,2016,24(11):109-119.
[11]范柏乃.我国城市居民生活质量评价体系的构建与实际测度[J].浙江大学学报(人文社会科学版),2006,36(4):122-130.
[12]张昆,迟国泰.基于相关分析-粗糙集理论的生态评价指标体系构建[J].系统工程学报,2012,27(1):119-128.
[13]迟国泰,曹婷婷,张昆.基于相关-主成分分析的人的全面发展评价指标体系构建[J].系统工程理论与实践,2012,32(1):112-119.
[14]陈洪海,迟国泰.基于信息显著性差异的绿色产业评价指标筛选模型及实证[J].运筹与管理,2014,23(6):193-204.
[15]赵志冲,迟国泰.基于似然比检验的工业小企业债信评级研究[J].中国管理科学,2017,25(1):45-56.
[16]韩伯棠,王莹.中国科技人力资源评价指标体系构建方法研究[J].北京理工大学学报(社会科学版),2006,8(6):31-35.
[17]Destrero A,Mosci S,Mol C D.Feature selection for high dimensional data[J].Computational Management Science,2009,6(1):25-40.
[18]顾雪松,迟国泰,程鹤.基于聚类-因子分析的科技评价指标体系构建[J].科学学研究,2010,28(4):509-514.
[19]周立斌,李刚,迟国泰.基于R聚类-变异系数分析的人的全面发展评价指标体系构建[J].系统工程,2010,28(12):56-63.
[20]赵宇哲,刘芳.生态港口评价指标体系的构建-基于R聚类、变异系数与专家经验的分析[J].科研管理,2015,36(2):124-132.
[21]Chen Honghai,Chi Guotai.Urban green development evaluation indicator system model based on clusteringrough set and application[J].ICIC Express Letters Part B:Applications,2015,6(10):2649-2654.
[22]王斌会.计量经济学模型及R语言应用[M].广州:暨南大学出版社,2015.
[23]郭亚军.综合评价理论、方法及应用[M].北京:科学出版社,2007.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |