The Feasibility of Cross-Validation as a Parameter Predictor for the Iterative Unfold Method
Abstract
Different applications of the Fredholm integral equation appear in varied fields of study. An application of particular interest to the Air Force arises when attempting to determine pulsed radiation spectra using measured data from underground nuclear effects simulations. The iterative unfold technique provides a means for approximating a solution to a system of Fredholm integral equations. This method consists of modifying a guess spectrum through a fit process using data collected from experiment. Next, the unfolded spectrum is smoothed to reduce undesireable artifacts that result from the fitting process. Finally, the entire iterative process is then repeated as necessary to provide an approximation to the exact incident spectrum. Presently, the iterative unfold method lacks an independent measure of how well the unfolded spectrum approximates the exact spectrum. Consequently, user judgement is necessary, resulting in possible data overfitting. Cross-validation is a method which selectively partitions the measured data into subsets. The subsets of data are used to predict omitted data. A cross-validatory loss statistic can be formulated and minimized to predict the optimum stopping point for the iterative unfold process. The iterative unfold technique was implemented into a computer program. Keywords: Theses.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1989
- Accession Number
- ADA220152
Entities
People
- Dennis J. Miller
Organizations
- Air Force Institute of Technology