| In today’s era of rapid development of science and technology,machine learning and data mining are gradually affecting people’s lives.Many scientific and technological products are applied on this basis,such as fingerprint or face recognition attendance equipment,automobile driverless,advanced intelligent robots and other technologies.These new technologies make people’s lives and work more convenient and efficient.Clustering analysis is an important application technology in the field of machine learning.The application of rough sets in clustering analysis is also a hot research area.This thesis introduces the commonly used measuring methods,and analyses the advantages,characteristics and application fields of various measuring methods.In view of the shortcomings of these measurement methods,this thesis proposes a measurement method based on neighborhood rough set theory with the idea of neighborhood rough set theory.At the same time,this thesis deeply analysis the characteristics,background and application scenarios of this method.In this thesis,the measurement method is applied to intelligent algorithm and clustering analysis to improve the performance of the original algorithm.The main research work of this thesis is as follows:(1)A measurement method of neighborhood rough sets is proposed.Firstly,the existing measurement methods are introduced through the theory of neighborhood rough sets.Then,the approximate accuracy and the approximate classification quality are obtained by using the approximate set of the neighborhood rough set,and the importance and roughness are calculated by the approximate accuracy and the approximate classification quality,thus the measurement method of the neighborhood rough set is proposed.Finally,the experiment shows that the measurement method can get better results.(2)The proposed measurement method based on neighborhood rough set theory is applied to particle swarm optimization(PSO)clustering algorithm,and a large number of experiments are used to verify it.Firstly,various improved measurement methods are analyzed and summarized,and representative measurement methods are selected for comparison.Then,experiments are carried out on real data sets and artificial data sets for different clustering evaluation indicators and different measurement methods.Finally,the clustering results are compared and analyzed in two aspects,including different algorithms using the same measurement method and the same algorithm using different measurement methods.(3)The proposed method is applied to density peak clustering algorithm.Firstly,the background and application scenarios of the measurement method are studied and analyzed.Secondly,the shortcomings of DPC algorithm are analyzed in detail,that is,when the DPC method uses clear neighborhood relationship to calculate local density,it can not recognize the neighborhood membership degree of different points from the distance,and it can not get ideal results when clustering unbalanced data and multi-peak data.Finally,the representative data sets are experimented with different clustering evaluation criteria. |
