摘要
信息融合方法是减少决策不确定性的有效途径和热点问题.本文研究模糊积分信息融合方法中的重要问题-模糊密度的确定方法,对其中两种典型的确定方法进行了细致的比较研究.基于公开而有效的13个UCI标准数据集,进行了成员分类器选择实验、不同融合方法比较实验等,并采用了描述分析、秩次分析、探测性显著性分析,最后与文献[4]中最优单分类器、文献[5]中Bagging,Boosting and random forests的最优融合结果进行了对比.结果显示,基于可能度的模糊积分方法优于基于可信度的模糊积分方法、优于文献[5]中最优融合结果;基于可信度的模糊积分方法与文献[5]中最优融合结果总体相当,优于简单平均融合方法,也优于文献[4]中最优单分类器.
Information Fusion is a valid way which can decrease the uncertainty of making decision, and is also a hotspot. The paper makes some work on a important problem about Fuzzy Integral, that is how to get the Fuzzy Density, and compares two typical means. Based on 13 UCI data set, this paper conducts the selecting experiment of membership classifiers and the compared experiment of several Information Fusion methods. It uses Descriptive Analysis, Rank Analysis, Exploringly Significant Analysis, and finally is compared with references 4 and 5. The result shows that the Fuzzy Integral method based on probability is better than the Fuzzy Integral method based on beliefs, is also better than the best results of single classifiers in references 4. The result also shows that the Fuzzy Integral method based on beliefs is nearly equal to the best results of fusion classifiers in references 5 in general, better than the average fusion method, and is also better the best results of single classifiers in references 4.
出处
《小型微型计算机系统》
CSCD
北大核心
2009年第2期283-288,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金重点项目(60234030)资助
国家基础研究项目(A1420060159)资助
湖南大学青年基(0723)资助
关键词
模糊积分
模糊密度
描述分析
秩次分析
探测性显著性分析
fuzzy integral
fuzzy density
descriptive analysis
rank analysis
exploringly significant analysis
