A Methodology for Defining the Number of Clusters and the Set of Initial Centers for Partitions Algorithms
Data clustering, Number cluster, Fuzzy C-Means, Initial centers
Data clustering consists of grouping similar objects according to some characteristic. In literature, there are several clustering algorithms, among which stands out the Fuzzy C-Means (FCM), one of the most discussed algorithms, being used in different applications. Although it is a simple and easy to manipulate clustering method, the FCM requires as its initial parameter the number of clusters. Usually, this information is unknown, beforehand and this becomes a relevant problem in the data cluster analysis process. Moreover, the design of the FCM algorithm strongly depends on the selection of the initial centers of the clusters. In general, the selection of the initial set of centers is random, which may compromise the performance of the FCM and, consequently, of the cluster analysis process. In this context, this work proposes a new methodology to determine the number of clusters and the set of initial centers of the partial algorithms, using the FCM algorithm and some of its variants as a case study. The idea is to use a subset of the original data to define the number of clusters and determine the set of initial centers through a method based on mean type functions. With this new methodology, we intend to reduce the side effects of the clusters definition phase, possibly speeding up the processing time and decreasing the computational cost. To evaluate the proposed methodology, different cluster validation
indices will be used to evaluate the quality of the clusters obtained by the FCM algorithms and some of its variants, when applied to different databases.