Abstract:
|
Fuzzy C means (FCM) has been widely used in biomedical research, however, its random initialization is a debating feature, and may lead to undesired steady states. Especially, the randomly selected initial centroids may locate far from the final converged centroids, therefore it may not be computationally efficient with large datasets. We propose a computationally enhanced Fuzzy clustering method (eFC) based on the data distribution density. The data density is used to guide the initialization of FCM, and improve its computational performance. The idea is to select the data points near the high-density area as the initial cluster centroids. Specifically, a directed-acycline-graph (DAG) structure is developed to describe the data distributions, and a prune-merge method is applied to locate the initial cluster centroids. Based on the real data and the simulation results, eFC seems to significantly improve the computation performance of FCM especially for large datasets.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.