Abstract: Most clustering methods need a pre-determined clustering number or a certain similarity threshold, which makes them dependent on heuristic knowledge. The X-means method tries to estimate the number of clusters but only converges locally. This paper presents a novel and simple clustering algorithm named as Auto-K after its descriptive parent-algorithm-K-means, though Auto-K theory can be generalized beyond certain given deriving algorithms. In Auto-K, the algorithm itself automatically selects a globally optimal clustering number for the involved population, by maximizing the clustering fitness and thus the clustering process can be said to be really dynamic and most accordant with human`s common sense in clustering.
Xiwu Han and Tiejun Zhao , 2005. Auto-K Dynamic Clustering Algorithm . Asian Journal of Information Technology, 4: 467-471.