A GLOBAL OPTIMUM CLUSTERING-ALGORITHM

The problem of clustering N patterns into m classes may be regarded as a combinatorial optimization, classification, or vector quantization. The goal of a clustering procedure is in most cases to minimize a cost metric. However, existing methods (such as the well-known K-means algorithm and its variants) are unable to avoid that clustering results are trapped into a local optimum. This paper develops a clustering algorithm based on a heuristic search strategy of artificial intelligence, which can determine the globally optimal classification. Furthermore, the algorithm is extended to promote the performance both in computation and in storage. An example on a real-world data set is illustrated. Also, the clustering problem with an unknown class number is dealt with in this contribution.