Double-quantitative Distance Measurement And Classification Learning Based On Tri-level Granular Structure Of Neighborhood System

The neighborhood rough set is an important tool for uncertainty analysis,and it’s closely related to granular computing.Therefore,the neighborhood system where the neighborhood rough set is located has become an important part of relevant information granulation and metric calculation.A tri-level granular structure(carrying the neighborhood granule,neighborhood swarm,and neighborhood library)has been constructed from the perspective of granular computing,and a granular computer system for knowledge learning has been established in neighborhood system.However,hierarchical exploration and related applications of tri-level granular structure of neighborhood system still have certain deficiencies.Therefore,this thesis supplements the correlation measures of tri-level granular structure of neighborhood system and extends them to classification learning.The relevant research of this dissertation mainly involves three aspects.(1)Based on tri-level granular structure of neighborhood system,the size measurement of bottom layer and middle group is considered to establish scale evaluation mechanism of highlevel neighborhood library.In order to complete the size measurement of neighborhood library,supplement and improve tri-level structure of neighborhood system at a higher level.(2)Based on relative distance and absolute distance of neighborhood granules in neighborhood system,the double-quantification technology is used to linearly fuse the relative distance and absolute distance of bottom neighborhood granules to construct a double-quantification distance;and all the three types of distances are promoted to both the middle swarm level and the top library level,so that the properties of related measurement are studied at three levels.(3)A classification algorithm is designed by using double-quantification distance that can effectively characterize the difference between neighborhood granules,and then doublequantification classifier KNGD(K-nearest neighbor classifier based on double-quantitative distance)is constructed.The decision table example analysis and the UCI data experiments verify the correctness of property of relevant measurement involved in this article,the results show that classifier constructed in this article is better than or balance the two existing classifiers,i.e.,relative classifier KNGR(K-nearest neighbor classifier based on relative distance)and absolute classifier KNGA(K-nearest neighbor classifier based on absolute distance).Thus the effectiveness of the algorithm KNGD built in this paper has been demonstrated.In general,this study has perfected three-layer granular structure of neighborhood system through theory,examples and experiments based on tri-layer granular structure of neighborhood system.The corresponding double quantitative integration and extension also provide in-depth knowledge measurement and effective classification learning,the related results have theoretical value and application significance.