A consistent technique for filling of unimodal clusters (Q1914135)

In cluster analysis, besides constructing a classification, we often encounter the problem of adding new observations to classes so that the filling technique remains consistent with the classification method with increase in ``training sample'' size and, furthermore, fewer computations are required for filling than for classification. A consistent filling technique based on the algorithm is proposed here for unimodal clustering. The method has linear computational complexity and is easily implemented on parallel computing systems with independent memories, e.g., on transputer networks.