Knowledge-based clustering: a semi-autonomous algorithm using local and global data properties

Cluster analysis is a heuristic technique used to reveal inherent groupings in data, but most modern clustering algorithms are highly data and person dependent. This paper presents a clustering technique that minimises the need for user-defined parameters and handles both single and mixed attribute type data sets. The algorithm is based on elements of rough set theory and uses a combination of local and global data properties to obtain meaningful clustering solutions. It is self-consistent in its approach to clustering; thus ensuring the same clustering solution when applied to the same data by different users. The results from a range of real-world and synthetic data sets are used to establish its performance.