Rough-Fuzzy Clustering Based on Adaptive Weighted Values and Three-Way Decisions

In the rough set-based clustering models, prototypes are iteratively updated by weighting the importance of core and boundary regions. The weighted valuewl is often pre-defined and fixed for all prototype calculations. In this way, the characteristics of data structures are not considered when assigning the weighted values. Some uncertainties may arise in the clustering processes, especially when the densities and sizes of different clusters are discrepant. In this study, an automatic mechanism for adaptively adjusting the weighted value w(l) is introduced which adheres to the distributions of approximation region partitions, and the uncertainties caused by the user-defined weighted values can be reduced. Based on the generated approximation region partitions of each cluster, an absolute boundary region is formed in which the samples are classified guided by the notion of three-way decisions. The validity of the proposed method is demonstrated by some benchmark data sets from UCI repository.