A dimensionality reduction approach to support visual data mining: Co-ranking-based evaluation

Kurzfassung

The content of Earth Observation archives continues to explosively grow generating the need to extract valuable knowledge from these repositories. A traditional approach for remote sensing image retrieval is to compute the similarity between the user's query and the archive using a query by example system. Additionally, effective data mining involves the user into data exploration, making use of his knowledge to initiate and validate new hypothesis. This paper brings into focus a visualization based approach to mining the EO data. This method aims to map the existing data correlations in the multidimensional information space to the spatial correlations revealed by the 3D space. The assessment of the results considers a single and global quality criterion, involving the number of the intrusions and extrusion to reveal the performance of dimensionality reduction methods.