Uncertain c-means clustering method with application to imprecise observations

Abstract

Cluster analysis is an essential method in machine learning, primarily used in situations with crisp data. However, data obtained in practice can be imprecise, forcing classic clustering methods to fail. Spurred by this constraint, this paper introduces an uncertain c-means clustering method, which employs uncertain variables to characterize imprecise observations based on the uncertainty theory. Specifically, we define a distance from an uncertain variable to a crisp vector and introduce an uncertain partition method. Additionally, according to the distance and partition method, an uncertain clustering is proposed. Finally, numerical experiments demonstrate the effectiveness of the proposed method.