Date of Award
9-1991
Document Type
Thesis
Degree Name
Master of Science
First Advisor
David A. Diener, PhD
Abstract
This exploratory study assesses the accuracy of backpropagation neural networks in predicting sortie generations, given pre-specified levels of air base resources. Single hidden layer networks and two-way interaction regression metamodels were fitted to simulated data previously generated by way of a factional design for ten factors at two levels, and subsequently tested (cross-validated) via an independent testing sample. It was determined that regression metamodels were generally superior in predicting unseen cases, while their network counterparts exhibited far better goodness-of-fit characteristics. The research consistently emphasizes that goodness-of-fit in no way necessarily implies goodness-of-prediction, in that different non-equivalent statistical measures are required to assess both these phenomena. In spite of their relatively poor performance in predicting the test sample used in this study, experimental results indicate that future research focused on applying neural network modeling techniques to sortie generation prediction and the identification of critical air base resources is warranted.
AFIT Designator
AFIT-GLM-LSM-91S-11
DTIC Accession Number
ADA246626
Recommended Citation
Dagg, James M., "An Exploratory Application of Neural Networks to the Sortie Generation Forecasting Problem" (1991). Theses and Dissertations. 8123.
https://scholar.afit.edu/etd/8123
Comments
The author's Vita page is omitted.
Presented to the Faculty of the School of Systems and Logistics of the Air Force Institute of Technology, Air University, in Partial Fulfillment of the Requirements for the Degree of Master of Science