Research On Data Effect-based Attribute Importance Measure Theory And Methods

With the rapid development of information technology,data from all walks of life is accumulated faster than ever.So methods of processing uncertain information have drawn a lot of public attention.Rough set theory,a mathematical theory used to process inaccurate,inconsistent and incomplete information,has long been not only an academic hot spot,but also widely applied in industry.Multi-attribute decision-making is a common problem in many fields,such as resource management,performance evaluation,information security and so on.Finding scientific and practical decision-making methods is essential to both academia and industry.For fuzzy roughness measurement: Based on analyzing the characteristics and shortcomings of the current fuzzy roughness measurement,this thesis puts forward the concept of membership effect function and establishes a fuzzy roughness measurement based on membership effect,abbreviated as FRM-BME.Then,several necessary and sufficient conditions of determining the value of FRM-BME are listed.Finally,a method for attribute reduction based on FRM-BME is proposed as an application of FRM-BME.This thesis further analyzes the characteristics and effectiveness of FRM-BME with specific cases.Both theoretical and experimental results show that FRM-BME not only has good intrinsic structural characteristics and interpretability,but also can perfectly integrate the fuzzy processing awareness into roughness measurement system.For attribute importance measurement: By studying the data system and the knowledge hidden in it,along with the contained degree of sets,this thesis proposes a data-delete method based on knowledge reliability,discusses the changing pattern of knowledge factor in the sub-data system,and puts forward an attribute importance measurement based on data effect,abbreviated as DE-AIM.Moreover,the thesis also discusses the value range and structural features of DE-AIM with theoretical analyses and calculations of examples.Finally,with the aid of specific cases and several common UCI data sets,there are some further analyses about the features of DE-AIM and the similarities and differences of other attribute importance measurements.Both theoretical and experimental results show that DE-AIM not only has good intrinsic structural characteristics and interpretability,but also can reflect different decision awareness with the adjustment of parameters,which indicates the promising prospect of DE-AIM in artificial intelligence,resource management,optimization of complex system,expert system and many other fields.