Font Size: a A A

Research On Clustering Methods For Uncertain Data Based On Feature Decomposition

Posted on:2018-12-13Degree:MasterType:Thesis
Country:ChinaCandidate:X Y SunFull Text:PDF
GTID:2348330515978432Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
With the changing technology environm ent,wireless sensor and data storage technology is developing rapidly,w e entered the real data age.The massiveness and diversity of data poses a great challenge for its m ining,and the emergence of large numbers of uncertain data furt her intensifies the difficulty of analysis.Uncertain data is widely found in the real world and it is difficult to a void.Due to the error of the physical instrument,the unreliability of network transmission,as well as human factors.The uncertainty of the data will affect the accuracy of the results of the data analysis seriously.so,in recent years,the research on uncertain data has been paid m ore and more attention.And how to reduce the impact of data uncertainty on data analysis has become one of the important research in the field of data mining.Data mining is an ef fective way to explore the p otential value of information,knowledge,and clustering is widely used in various fields of society as an important part of data m ining.As a basic method of m achine learning and pattern recognition,it divides the data through the unsupervised learning model,and can obtain the valuable information effectively.Therefore,this paper will explor e the ef ficient clustering methods for uncertain data.The existing clustering m ethods of uncertain data are mostly based on the clustering method of deterministic data.Although the improved methods can enhance the effect of clustering uncertain data to a certain extent,there are still many problems,For example,the result of clustering is seriously affected by uncertainty,and the applicability of the methods is low.In view of t he above problems,this paper has carried out targeted research,the main work is as follows:Firstly,this paper elaborates the signi ficance of the clustering m ethod for the uncertain data,summarizes the outstanding research results at hom e and abroad,classifies the ways to im prove the clus tering method for the uncertain data,and analyzes the existing problems.Secondly,this paper proposes an cluste ring method for uncertain data based on feature decomposition for the existing problem s of the improved methods.First of all,the basic th eory of uncertain data is su mmarized,and the basic principles of the method are explained,such as the principl e of the covariance structure and the sharpening process,etc.At last,describe the detailed process of the method.The method use the n atural potential association of the data.According to the covariance structure,using the data analys is method based on spectral d ecomposition,we can get the sharpening data.Then,do clustering analys is of the sharpening data,in order to achieve better results in clustering uncertain data.And on this basis,a fuzzy clustering method of uncertain data based on feature decomposition is proposed.Thirdly,this paper compares the proposed clustering method for uncertain data with the classical K-m eans method,compares the proposed fuzzy clustering m ethod for uncertain data with the classical fuzzy c-m eans method,and verifies the effectiveness of the m ethods.The experimental process includes the selection of the clustering evaluation index a nd experimental data set,th e construction of uncertain data and so on.Finally,according to the result s of comparison experiments,it is proved that the m ethods can more effectively clus ter the uncertain data,and the data after sharpening has wide applicability and practicability.
Keywords/Search Tags:Uncertain data, feature decom position, clustering, covariance structure, data sharpening
PDF Full Text Request
Related items