The Optimal Scale Selection For Multi-granular Labeled Decision Systems

Rough set theory,proposed by Polish mathematician Pawlak in 1982,is a mathematical theory to analyze and deal with incomplete data.It uses only internal knowledge,avoids external parameters,and does not rely on prior model.Its basic idea is to unravel a set of decision rules from an information table via knowledge reduction process which determines the necessary and sufficient attributes constituting the rules for classification.With the development of rough set theory,it has been successfully applied in many real life fields,especially in the field of agricultural informatization in recent years.Granular computing is an important method for data mining and knowledge representation.It is a very active research direction in the field of artificial intelligence,especially in intelligent information processing.It simulates human thinking mode and uses granules as basic units of computation to deal with large-scale complex data and information to establish an effective computational model as the goal.Granular computing is characterized by seeking to solve complex problems with an approximate method at a suitable granularity level.In real life,people often from the different levels of granularity to observe the object or process data,the choice of the appropriate decision-making system and the corresponding decision-making rules in different levels of granularity is a very important research topic in intelligent information processing.In order to solve the problem of knowledge acquisition in multi-granular labeled decision system,this dissertation proposes a method of selecting optimal granularities in multi-granular labeled decision systems.The concepts of information systems,multi-granular labeled information systems,and multi-granularity labeled decision systems are first introduced.Optimal granularity selections for multi-granular labeled decision system with different meanings are then discussed.Eight types of optimal granularities in multi-granular labeled decision systems are defined and their relationships are further examined.It is proved that there are in fact four types of different optimal granularities.Finally,approaches of attribute reduction and ruleextraction in the sense of four types of optimal granularity in multi-granular labeled decision systems are,respectively,presented.