Research On Clustering Of Uncertain Data

Uncertain data is a forthcoming area in data mining, with numerous applications in wireless sensor networks (WSN), radio frequency identification (RFID), etc. This paper introduces the reasons causing uncertain and data types, then summarizes existing algorithms for clustering on uncertain data. After all of these, this paper illustrates that all the works on clustering on uncertain data which are extended on traditional clustering method.Since the clustering problem on numeric data sets can be formulated as a typical combinatorial optimization problem, and it has been proved to be NP-hard. The backbone (the shared common parts of all optimal solutions)as a useful method can solve the NP-hard problems. However, it is hard to gain the backbone, and replaces with the approximate backbone by using the intersection of local optimal solutions. This paper provides a new algorithm ABAUDC as a framework which is based on approximate backbone.The new algorithm includes three models, and firstly, it uses CKMeans algorithm for the initial clustering, and it reduces the complexity of computing expected distances. Secondly, approximate backbone is generated by intersection of local optimal solutions, and the search space can be dramatically reduced by fixing the approximate backbone, then it reduces search space can be efficiently searched to find high quality solutions. Thirdly, the semi-supervised UKMeans algorithm is given in ABAUDC to cluster with constraints.The experiments on UCI datasets such as HabermanSurvival, Wine, Soybean and Glass prove that ABAUDC algorithm can improve clustering precision than UKMeans. And it proposes that ABAUDC algorithm is a new method with flexibility and simplicity on clustering for uncertain data.