Research On Local Data Analysis Method Based On Decision-theoretic Rough Set

In the coming era of artificial intelligence,the importance of data is getting bigger and bigger.In recent years,the development of new technologies,such as large data,cloud computing,and Internet of things,often can not be separated from data mining.As one of the most important methods of intelligent information processing,rough set theory is a mathematical tool proposed by Poland scholar Z.Pawlak in the 80 s of last century to deal with uncertain problems.However,with the development of society,the diversity and variability of the data source,and the complexity of the source and structure,the classical rough set theory is no longer suitable for the needs of reality.How to discover from large and complex data in deeper information and implicit knowledge becomes more challenging.In this paper,we start from the local thought and then we explore the problem of rough set theory modeling and attribute reduction in the aspects of the construction of multi feature space and the sensitivity of testing cost.The main research results are as follows:1.Local data analysis of decision rough sets.Compared to the classical rough sets,decision rough set model takes cost into account,which brings some new challenges for solving attribute reduction in rough set.Some attribute reduction methods of decision rough set have been put forward,however,these standards of reduction are based on all decision classes,more stringent constraints.To solve this problem,from a local perspective,the idea Local attribute reduction is been proposed.The result based on heuristic algorithm shows that compared with the reduction for all decision classes,Local attribute reduction can get more positive domain rules and reduce the number of attributes.2.Rough data analysis based on multi-feature space.Rough data analysis based on multi-feature space.From the local idea,it is considered that each decision class itself may have its own unique characteristics.From the inspiration of multi-label learning,a multi-feature space which can reflect the basic characteristics of information system is constructed by clustering analysis.The definition of upper and lower approximation and approximate quality based on decision rough set is given.Through the experimental analysis,it is found that the structure of multi-feature space can effectively reduce the uncertainty of the information system and improve the classification performance.3.Cost-sensitive decision rough set model.It is an innovation in this paper to introduce the test-cost into the decision rough set model and design the corresponding attribute reduction algorithm.When it comes to the incomplete information system,the test-cost sensitive decision model of rough set constructed in this paper is also the generalization of rough set model based on tolerance relation,this will be more close to the reality.Meanwhile,on the basis of generalization modeling of decision rough set model,considering that the traditional heuristic algorithm does not introduce the test costs into it,this paper proposes a cost sensitive ?-cut quantitative decision reduction algorithm.