Research And Application Of Evidential Clustering Algorithm With Applications

With the rapid development of information technology,the total amount of data is growing exponentially.Companies rely on their powerful storage capability to continuously collect,organize and analyze data,in order to mine valuable information.A large amount of data is stored in different data sites and various types of servers.Due to security,privacy or other technical reasons,companies are reluctant to share data and only want to exchange information at nondata levels.In order to make better use of different levels of information and reveal the internal information structure at local data sites,clustering algorithms based on collaborative mechanisms have been proposed.The basic idea of collaborative clustering is to first run a clustering algorithm independently at each data site,and then interact by exchanging the local structure information of each data site to reveal the potential common underlying structure of different data sites.The evidential clustering algorithm is based on the concept of credal partition in evidence theory,which expands the traditional hard partition,fuzzy partition and possibilistic partition algorithms to better understand the internal structure of data.In the framework of evidence theory,this paper first proposes the concept of collaborative evidential clustering by introducing collaborative mechanism to enhance evidential c-means algorithm to explore the deeper structure information of each data site.Firstly,the collaboration mechanism under the framework of evidence theory is established among the credal partition matrices of each data site to meet the data confidentiality requirements.Secondly,considering the problems of excessive information interaction and insufficient information interaction,we design single-step and multi-step collaborative evidential clustering algorithms respectively.The number of steps in multi-step collaborative evidential clustering algorithm is effectively controlled by the global structural similarity index.Thirdly,for the collaborative intensity coefficient that balances the local data information and the structure information of collaborator,we give two indicators of inter-cluster similarity and cluster relative distance for auxiliary guidance,which are convenient for user-defined setting.Finally,in order to prove the effectiveness of collaborative evidential clustering algorithms and the rationality of indexes,the collaborative effects are investigated more comprehensively at the global level and local level through the simulated data set and the three real data sets in the UCI machine learning repository.