Research On Possibilistic Fuzzy Clustering Algorithm With High-Density Points

Fuzzy clustering algorithms are usually data-driven.Recently,knowledge has been introduced into these methods to form knowledge-driven and data-driven fuzzy clustering algorithms,which has make a new breakthrough in the field of cluster analysis.In those algorithms,some knowledge hints are merged with the original data,and the partition matrix is obtained by the knowledge-driven clustering algorithms.Finally,the clustering result with or without knowledge hints is obtained based on the partition matrix.The whole process is very similar to the way people classify unknown objects,and the results of algorithms are more in line with reality.However,knowledge-driven clutering algorithms have the problems of sensitity to the initialization of clustering centers,artificial acquisition for knowledge points and the number of cluters need to be given in advance.To this end,this dissertation will study and improve the knowledge-guided fuzzy clustering algorithm for these problems.The research work includes the following aspects:(1)Aiming at the clustering center initialization sensitive problem of fuzzy clustering and the knowledge points extraction problem of knowledge-driven clustering,Hyperspher Density-based Clustering Center Initialization(HDCCI)method and Density Knowledge Points Extraxtion method are proposed.The HDCCI algorithm can automatically get the C initial clustering centers in the center of the dataset structure.Knowledge-drive clustering algorithm take the highest dentity point as the viewpoint,which is also one of the prototypes.The viewpoint plays a guiding role in the clustering process.The DKPE algorithm can obtain several data points with obvious higher density.The extracted high-density points can be used as the knowledge hints of the knowledge-driven possibilistic fuzzy clustering algorithm to make the algorithm acquire more accurate prototypes.(2)In order to enhance the anti-noise performance of fuzzy clustering algorithms,this dissertation put forward Density Viewpoint-induced Possibilistc Fuzzy C-Means(DVPFCM)algorithm.The algorithm takes the high-denstiy point obtained by the HDCCI method as a new viewpoint and integrates it into the possibilistic clustering algorithm.The integration of the viewpoint knowledge makes the DVPFCM algorithm get the ideal clustering result faster and has stronger robustness.(3)Proposed High-density Points-driven Adaptive Possibilistic C-Means(HPAPCM)algorithm in this dissertation can automatically determine the number of clusters.First,according to the high-density points extracted by the DKPE algorithm,the initial value of the number of clusters can be determined to be 2 times,which is over the true clusters number,and the high-density points are included in the objective function of the HPAPCM algorithm to guide the iterative update of the cluster center and the membership matrix.In the iterative process,the HPAPCM algorithm can adaptively eliminate the isolated cluster centers,and gradually get the number of cluster centers close to the actual number of clusters.In the HPAPCM algorithm,the high-density point is not directly used as the clustering prototype,but actually plays a guiding role such that the guided algorithm automatically obtains more reasonable clustering prototypes.