Interactive visualization and analysis of hierarchical neural projections for data mining

Dimensionality reducing mappings, often also denoted as multidimensional sc aling, are the basis for multivariate data projection and visual analysis i n data mining. Topology and distance preserving mapping techniques-e.g., Ko honen's self-organizing feature map (SOM) or Sammon's nonlinear mapping (NL M)-are available to achieve multivariate data projections for the following interactive visual analysis process. For large data bases, however, NLM co mputation becomes intractable. Also, if additional data points or data sets are to be included in the projection, a complete recomputation of the mapp ing is required, In general, a neural network could learn the mapping and s erve for arbitrary additional data projection. However, the computational c osts would also be high, and convergence is not easily achieved, In this wo rk, a convenient hierarchical neural projection approach is introduced, whe re first an unsupervised neural network-e.g., an SOM-quantizes the data bas e, followed by fast NLM mapping of the quantized data. In the second stage of the hierarchy, an enhancement of the NLM by a recall algorithm is applie d. The training and application of a second neural network, which is learni ng the mapping by function approximation, is quantitatively compared with t his new approach. Efficient interactive visualization and analysis techniqu es, exploiting the achieved hierarchical neural projection for data mining, are presented.