A Hybrid Fuzzy Clustering Algorithm And Its Application

With the coming of information age, people are in a state of data explosion and being lack of knowledge. Therefore, data mining and processed become increasingly important and a number of algorithms dealing with various data invented with it.Since Zadeh proposed fuzzy theory, fuzzy clustering analysis has developed quickly and has been widely used in many fields. There exists many fuzzy cluster analysis methods and the most popular is FCM ( fuzzy c -means) clustering method among the objective function-based clustering methods. FCM method achieves the division of a given sample set through iterative optimization to the objective function. The current fuzzy cluster analysis studies based on the objective function are based on this basic theory. FCM clustering can give each sample belonging to a cluster's membership and even for the clear classification of variables it also acquires a more satisfactory result.However, it has some shortcomings: the class number of clustering does not automatically determine, when being used the effective clustering criteria must be determined; Class center's location and characteristics do not always be known in advance but it must be generated by randomly initialized; Convergence speed is slow and easily falls into local minimum points and could not get the best optimal classification, it is more sensitive to the initialization; The computing overheads for clustering large data sets; The algorithm is more sensitive to the noise data in many cases.In fact transitive closure method, which is based on fuzzy equivalence relations, is the classification of transitive closure. It easily leads to so-called "transmission bias" when transforming the process of fuzzy similar matrix and classification results does not match the actual situation seriously, and the outcome of this method are a series of classification. The FCM method based on the fuzzy partition requires the initial division of a given matrix must have some basis and identifying the number of cluster classes in advance. A hybrid fuzzy clustering algorithm is presented in this paper with the combination of two methods. First, with the transitive closure method obtaining a series of classification and the next thing is introduction to F - statistics, according to F - the size of the statistics to get a classification, with the greatest F - the statistics class classification corresponds to a number of categories as the numberc of FCM clustering, the cluster centers in such categories as the initial cluster center of the FCM algorithm.In this paper the Matlab7.0 programming is used to combine with the fuzzy mathematics, mathematical statistics, optimization theory and pattern recognition, and a new cluster analysis method is proposed that do not need to pre-design number of clusters and is able to reduce its sensitivity to initialize. On the other hand, the hybrid fuzzy clustering algorithm can also be applied to the analysis of agriculture production.