Study On Attribute Reduction Criteria And Information Loss Of Attribute Reduction Based On Rough Sets

With the rapid development of the big data period,the amount of data in the database is increasing.The characteristics of these data are:Volume,Velocity,Variety,Value,Veracity.For these diverse and massive data sets,data reduction is the primary issue for the big data.How to take effective techniques to get the most valuable information from large amounts of data quickly,so that the data sets to achieve the effect of reduction has become a hot topic.Rough set theory is a mathematical tool used to deal with inaccurate,incomplete and fuzzy information.It has strong qualitative and quantitative analysis capabilities,can effectively express uncertain or inaccurate knowledge,and can use uncertain,incomplete empirical knowledge for analysis and reasoning.Attribute reduction is the core content of rough set theory research.Researchers have proposed many methods for attribute reduction criteria.They all have one thing in common.One thing is that attribute reductions only keep classification unchanged and information unchanged under certain reduction criteria.This article will explore and experiment with the concept of "attribute reduction without loss of information".Based on the information theory,quantitative analysis of the information loss caused by attribute reduction,and with examples to explore the impact of attribute reduction information loss on data classification.This paper proves that people have long misunderstood the information loss of attribute reduction,which lays the foundation of information theory for further study of attribute reduction and classification of rough set.The main innovations in this article are as follows:1.A conditional attribute reduction rule for rough sets is proposed.This paper comprehensively analyzes and summarizes the attribute reduction criteria existing in rough sets,and summarizes the general rules of attribute reduction.Based on this general rule,a new method of attribute reduction guidelines is proposed,and examples are used to verify its rationality and effectiveness.2.It proposes the measurement and calculation method of information loss in rough set attribute reduction.From the probability theory,information theory point of view of two quantitative observe,analyze different conditions simple guidelines about the properties,before and after comparison of attribute reduction occurs.Information entropy is used in thermodynamics to represent the average amount of information in a message that excludes redundancy.In this paper,we will use the concept of information entropy in thermodynamics to explore the information loss of Pawlak's rough set attribute reduction,and then give the measurement method and calculation formula of attribute reduction information loss.3.To explore the relationship between attribute reduction information loss and classification accuracy is the most key.The potential impact of experiments take ten fold cross-validation method,using the average of ten groups of data finally calculated as the loss of different information to explore various attributes of the data set reduction of data classification.4.It focus on exploring the application of attribute reduction information loss in a multi-granular rough set.In this chapter,the joint entropy is used to measure the information loss of multi-granularity rough set attribute reduction.Compared with information entropy,the participation of decision attributes is increased,and the data division is more precise and detailed.According to the characteristics and properties of the loss of granularity information,an attribute reduction algorithm for granularity information loss is designed.