Attribute clustering with unknown cluster numbers

In this paper, we try to select features based on attribute clustering without knowing the exact cluster numbers in advance. A similarity measure for a pair of attributes is first described, and an attribute clustering approach based on the CAST algorithm is then proposed to group the attributes into adequate number of clusters. The representative attributes found in the clusters are thus used for classification such that the whole feature space is greatly reduced. If the values of some representative attributes cannot be obtained from current environments for inference, some other possible attributes in the same clusters can also be used to achieve approximate inference results.