The problem of clustering is that of separating a data set into a number of groups (called clusters) based on some measure of similarity. The goal is to find a set of clusters for which samples from different clusters. Often a local prototype is also produced, which characterizes the members of a cluster as a group. The structure of the data is then inferred by analyzing the resulting clusters (and prototypes) by domain experts. Since the clustering is usually used for interpretation, the similarity measured in the clustering process is subjectively etermined. A common strategy is to minimize the squared error as is done in vector quantization, Fuzzy clustering methods seek to find fuzzy partitioning by minimizing a suitable (fuzzy) generalization of the squared loss cost function. The goal of minimization is to find centers of fuzzy clusters, and to to assign fuzzy membership values to data points. The resulting algorithms are similar to traditional vector quantization/crisp clustering methods (i.e. k-means).
展开▼
