Sampling and Subsampling for Cluster Analysis in Data Mining: With Applications to Sky Survey Data

This paper describes a clustering method for unsupervised classification of objects in large data sets. The new methodology combines the mixture likelihood approach with a sampling and subsampling strategy in order to cluster large data sets efficiently. This sampling strategy can be applied to a large variety of data mining methods to allow them to be used on very large data sets. The method is applied to the problem of automated star/galaxy classification for digital sky data and is tested using a sample from the Digitized Palomar Sky Survey (DPOSS) data. The method is quick and reliable and produces classifications comparable to previous work on these data using supervised clustering.