HOW TO CHOOSE A REPRESENTATIVE SUBSET FROM A SET OF DATA IN MULTIDIMENSIONAL SPACE

Given a set of N points in multi-dimensional space, it may be necessar y to choose a subset of n representative points. For example, in clust ering problems, it is necessary to choose a few seed points around whi ch the cluster may grow. This problem may be posed as that of choosing one out of each k data when right perpendicular N/n left perpendicula r=k. In our proposed method, the data points are ordered in decreasing magnitude of density. The datum toping the ordered list is chosen and its k-1 nearest neighbours are deleted from the ordered list. From th e remaining data, the one currently toping the list is chosen. The pro cess is repeated till the data are exhausted. The problem of more gene ral choice of n is also addressed.