The objective of clustering algorithms is to group similar patterns in one class and dissimilar patterns in disjoint classes. This article proposes a novel algorithm for fuzzy partitional clustering with an aim to minimize a composite objective function, defined using the Fukuyama-Sugeno cluster validity index. The optimization of this objective function tries to minimize the separation between clusters of a data set and maximize the compactness of a certain cluster. But in certain cases, such as a data set having overlapping clusters, this approach leads to poor clustering results. Thus we introduce a new parameter in the objective function which enables us to yield more accurate clustering results. The algorithm has been validated with some artificial and real world datasets.
展开▼