ABARC: An agent-based rough sets clustering algorithm

Clustering is an important task in pattern recognition with many applications in natural sciences and healthcare. However, in practical scenarios, it is often the case that the data cannot be easily separated into well distinguished groups for several reasons like: the shape of clusters, the presence of outliers, or the overlapping clusters problem (instances that may belong to more than one cluster). In order to handle such issues, we propose an agglomerative clustering approach which identifies instances that may belong to more than one cluster and clearly separates the outliers form the rest of the instances by integrating concepts from rough sets theory. The whole grouping and regrouping process is driven by software agents executing in parallel. Our approach is computational friendly and experiments on standard data sets indicate its advantages.