Actually, that's not the motivation of density-based clustering, the example you gave can be solved by measuring the distance in a proper way.
One characteristic of the density-based clustering is that it can leverage the local information of each data point to cluster the data to obtain the clustering with arbitrary shape. Another characteristic is that the density-based clustering is able to discover the data points that does not belongs to any cluster, and mark them as outlier. Although the spectral clustering that clusters the data based on the similarity matrix can partially achieve the first characteristic, but the built-in K-means clustering inside the spectral clustering make the second characteristic not so easy to achieve. 2013/5/8 Ted Dunning <ted.dunn...@gmail.com> > On Wed, May 8, 2013 at 2:33 PM, yu lee <leeyufam...@gmail.com> wrote: > > > e.g., how to cluster two buildings > > which are quite close to each other in terms of their euclidean distance, > > yet with a river in between them? > > > > Define distance in SVD space based on how people move. > > Next problem? > -- ------ Yexi Jiang, ECS 251, yjian...@cs.fiu.edu School of Computer and Information Science, Florida International University Homepage: http://users.cis.fiu.edu/~yjian004/
