Data analysis and feature selection for discrimination are, for the most part, very manual and time consuming tasks which are non-optimized. While an experienced algorithm developer will have reasonably good success in the selection of robust features, it is impossible to manually analyze the many sets of actual test and simulated data and select the truly optimal set of features for discrimination. The process requires the algorithm developer to spend weeks or months in the analysis/selection of features. Another challenge is the viewing/analysis of multidimensional features simultaneously. Standard visualization techniques include scatter plots and histogram mapping are dimensionally limited. There is an inherent need to quickly process large sets of multidimensional data/features (across a wide range of scenarios, lethal, non-lethal, and countermeasures) and automatically select the optimum set of features. EWE proposes a unique software tool for data analysis, feature selection, and visualization. This integrated tool will allow the algorithm developer to process a wide range of data sets (visible, infrared, radar, ladar); incorporate classical and modern techniques for feature selection/clustering; and provide a variety of visualization options and will directly translate into a more robust discrimination algorithm with improved performance. Anticipated Benefits/Commercial Applications: The technology proposed here will have wide range of applications for all projects involving feature selection and discrimination in Government as well as private commercial sector. Potential DOD applications include, project Hercules, THAAD, GMD.
Keywords: Feature Selection, Discrimination, Data Analysis , Clustering, Multi Dimensional Visualization
