An Approach to Perform Local Analysis on Multidimensional Projection

In the context of Visualization, Multidimensional Projection techniques are employed to show similarity relations among instances of a multidimensional dataset. Distinct projection techniques use different approaches to perform the dimensionality reduction and, consequently, different metrics are employed to assess projection quality according to similarity and structures preservation. Usually, quality measures are computed from the whole projection, what can impair a specific evaluation. This work presents a novel approach to perform evaluation on multidimensional projections, in which clusters of instances are selectively evaluated and compared to the whole projection. The proposed approach has shown to be effective on evaluating projections and it offers a way to apply techniques to enhance poor projected areas.