The Study Of Attribute Reduction Algorithm With Cost-sensitive In Rough Set Theory

In recent years,with the rapid development of artificial intelligence and knowledge discovery,the application of computer technology in various fields is increasing day by day.The amount of data increases dramatically,with these characteristics of high dimension,multi-category and complex structure.It is of great research significance and application value to process and mine these data and obtain valuable knowledge from it.To effectively process and analyze incomplete and uncertain data,rough set theory,as a mathematical tool,provides a good theoretical support for data mining research.At the same time,cost sensitive learning,whose aim is to obtain a minimum set of total cost attributes,is one of the ten challenging problems in the field of data mining.Therefore,based on the cost sensitive rough set theory,this paper proposes two attribute reduction algorithms based on sensitive learning from the perspective of test cost and misclassification cost for different data types in the application field.Firstly,a cost sensitive attribute reduction algorithm for incomplete neighborhood data is proposed.As cost sensitive attribute reduction is a natural extension of classical attribute reduction,its research has more practical application significance.At present,the research of cost sensitive attribute reduction mainly deals with nominal data in complete decision tables.But in practical application,due to the absence and omission of information,the data presented is incomplete.Therefore,for the above problems,this paper constructs an attribute reduction algorithm for incomplete neighborhood data based on feature test cost and decision misclassification cost.Firstly,the algorithm calculates the neighborhood granularity of incomplete neighborhood data,and then calculates the attribute reduction result based on the heuristic strategy of kernel attributes according to the newly proposed measure method of attribute importance based on test cost and misclassification cost.Finally,the feasibility and rationality of the algorithm are illustrated by theoretical analysis and examples analysis.Secondly,a cost sensitive attribute reduction algorithm for incomplete mixed data with variable precision is proposed.Since multiple data types coexist in practical applications,the acquisition of data needs to pay a certain cost and prize.Therefore,it is especially important to study the attribute reduction of incomplete mixed data under cost-sensitive learning.In order to solve this problem,a variable precision rough set model is introduced,and a cost-sensitive variable precision incomplete mixed data attribute reduction is proposed.The algorithm firstly finds the global neighborhood granularity and positive domain according to the threshold and variable precision,and finds the positive domain under each attribute,and then solves it according to the newly defined test cost based on test cost and misclassification cost attribute.The importance of the attribute is calculated by the attribute reduction algorithm based on the greedy search strategy.Finally,through case analysis and experimental comparison,compared with the algorithm without considering the cost,the algorithm can obtain the attribute reduction result with low total cost,which further demonstrates the effectiveness of the algorithm.Finally,a comprehensive summary and prospect on the algorithm and theoretical knowledge of this paper is made.