Array Component Analysis with Application to Remote Sensing Atmospheric Science Data

• Activity Number: 75
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Irina Kukuyeva*+
• Keywords: Dimension Reduction ; ICA ; Remote Sensing ; data mining

Abstract

Multivariate data sets are common in statistics, though it is not always the case that all variables and their interactions can be considered simultaneously due to computational limitations. Therefore, analysis of such a wealth of information needs to find the optimal tradeoff between accuracy and volume, since the more data we have, the more detailed it is, but the harder it is to analyze. Because of this existing tradeoff, researchers turn to dimension reduction techniques, which simultaneously reduce the data volume while keeping most of the information in the data set intact. We extend the current toolbox of these techniques to handle non-normally distributed data sets with three or more modes, which can be thought of as a generalization of Independent Component Analysis (ICA). We assess the success of the approach, called Array Component Analysis (ACA), by applying the new methodology to remote sensing.