Date of Award
3-2008
Document Type
Thesis
Degree Name
Master of Science in Operations Research
Department
Department of Operational Sciences
First Advisor
Kenneth W. Bauer, Jr., PhD
Abstract
The purposes of this research were: (1) the modeling of a CID situation and (2) the search for robust and controllable input variable settings. The inputs were defined as controllable and noise variables and the confusion matrices in ROC theory were adapted to act as controllable factors. In this research a simple virtual battlespace representation is employed. The experimental results of the CID system are summarized by a posterior confusion matrix and throughout the confusion matrix analysis we can obtain all various types of data such as accuracy, error cost, error rates, and so forth. To find the optimal parameters three evaluation techniques were applied: (1) Linearly constrained discrete optimization, (2) Taguchi’s S|N ratio method and (3) Robust parameter design with a combined array. The results are compared and contrasted across different objective functions. In conclusion, if we consider the diverse characteristics of CID, the simulator needs to focus on finding the controllable parameter that yields the maximum accuracy value. This is because the minimum cost is typically accomplished at the point of maximum accuracy and the cost approach is very subjective depending on the decision maker and battlefield situation. In addition, the most preferable evaluation method is RPD with a combined array due to its superior performance outside of the design space. In the final analysis, we need a detector/classifier that has good performance to minimize error costs and maximize label accuracy.
AFIT Designator
AFIT-GOR-ENS-08-11
DTIC Accession Number
ADA482836
Recommended Citation
Kim, TaeHo, "Combat Identification Modeling Using Robust Optimization Techniques" (2008). Theses and Dissertations. 2811.
https://scholar.afit.edu/etd/2811