An Adaptive Neighborhood Retrieval Visualizer

摘要

We propose a novel adaptive version of the Neighborhood Retrieval Visualizer (NeRV). The data samples' neighborhood widths are determined on the basis of the data scattering in the high-dimensional input space. The scattering of input data is measured using the inner-cluster variance quantity, obtained as a result of the preliminary data clustering in the input space. The combination of the pre-clustering and the subsequent NeRV projection can be recognized as a hybrid approach. The experimental study carried out on two different real datasets verified and confirmed the effectiveness of the introduced approach and the correctness of the theoretical claim of the paper.