Multidimensional Projections Analysis Using Performance Evaluation Planning

Abstract

Among the challenges within the big data era, analyses of high multidimensional data are still an open research area. As a result, several multidimensional projections techniques have been developed to reduce data dimensionality, becoming a powerful visualization and visual analytics tool. In order to ensure the projection quality, it is necessary to assess the lower-dimensional embedding by using different datasets configurations as input and analyzing evaluation metrics. However, it is not always clear to the user how the number of dimensions, instances or clusters, called factors, can affect the projection mapping and its quality regarding different projection techniques and assessment metrics. This work aims to address how much the factors affect each response variable through performance evaluation planning. We present an evaluation approach that carries out a more in-depth multidimensional projection analysis supported by the factorial design. Through the comparison of two multidimensional projections, it allows a better understanding of how distinct dataset properties can influence on quality metrics results.