Agglomerative hierarchical clustering for nonlinear data analysis

Clustering methods based on nonparametric density estimation define a cluster as the basin of attraction of a local maximum (mode) of the density function by mode seeking with the mean shift procedure. The efficacy of mean shift analysis has been demonstrated in many problems. However, one of the limitations of the mean shift procedure is that it involves the specification of a bandwidth parameter. This requires a prior knowledge of a range of variable bandwidths, strong dependent on applications. To avoid this disadvantage, we propose a new method based on agglomerative hierarchical clustering concept that begins with small clumps using a cell-based clustering scheme, and then successively merges neighboring clumps together until a stopping criterion is satisfied. The proposed algorithm is successfully tested on complex clustering examples. Some experimental results of comparing the proposed method with the classical mean shift show that our algorithm is robust and efficient.