A Study On The New Methods For Clustering Analysis

Clustering analysis plays an important role in the research of pattern recognition. A series of new algorithms and new concepts are presented in this dissertation to surmount the defects of the existing clustering algorithms:Firstly , a new competitive learning algorithm is developed to overcome the shortcomings of the competitive learning algorithms available . The new algorithm has faster convergence speed and higher clustering accuracy in comparison with the algorithms present, and is a further advance in the competitive learning theory.Secondly, a new explanation of the membership degree is discussed. The new explanation is of great importance in understanding the advantages and disadvantages of the fuzzy clustering , hard clustering and possibilistic clustering.Thirdly , a new soft clustering algorithm called the rival checked fuzzy C-means algorithm is proposed to improve the convergence speed of the fuzzy C-means algorithm.Fourthly , the fruits achieved in the research of the two types of the fuzzy Cspherical clustering algorithms fill the gap of the research in the fuzzy C-spherical clustering.Fifthly , a new algorithm is derived to counter the shortcomings of the Euclidean distance-based fuzzy C-spherical shell clustering algorithms. The new algorithm has excellent properties in both the convergence speed and clustering accuracy.Sixthly , the distance modification-based fuzzy C-spherical shell clustering algorithm is provided to improve the convergence speed.Seventhly , a new concept that the missing of the rules has an effect on the properties of the fuzzy associative memories is proposed to refute the one-sided concept proposed by the author of the fuzzy associative memories.Finally, a new clustering method for extracting the rules of the fuzzy associative memories gets rid of the shortcomings of the existing clustering algorithms and is proved to be a good methodIt抯 can be drawn from the above that the fruits achieved in the dissertation enrich the theory of the pattern recognition and possess significant values in the application.