Volume 6 Number 4 (Oct. 2016)
Home > Archive > 2016 > Volume 6 Number 4 (Oct. 2016) >
IJAPM 2016 Vol.6(4): 150-164 ISSN: 2010-362X
doi: 10.17706/ijapm.2016.6.4.150-164

Analysis of How Well Regression Models Predict Radiation Dose from the Fukushima Daiichi Nuclear Accident

Stephen U. Egarievwe, Jamie B. Coble, Laurence F. Miller
Abstract—The 2011 Fukushima Daiichi nuclear accident in Japan resulted in the release of radioactive materials into the atmosphere, the nearby sea, and the surrounding land. Based on the International Atomic Energy Agency (IAEA) Convention on Early Notification of a nuclear accident, several radiological data were collected on the accident. Among the radioactive materials monitored, are I-131 and Cs-137 which form the major contributions to the contamination of drinking water. The radiation dose in the atmosphere was also measured. This study focused on how well regression models predict radiation dose from the following predictor variables: I-131and Cs-137 concentrations in drinking water, radiation monitoring locations, and distance and direction of monitoring points from the accident location. The analysis covered 1) the correlations between the radiation dose and the predictor variables, and 2) how well simple regression methods could predict the radiation dose. The modeling techniques investigated include linear regression, principal component regression (PCR), partial least square regression (PLS), ridge regression, and locally weighted regression. The Venetian Blinds method was used to divide the data into training, test, and validation datasets. The concentrations ofI-131 and Cs-137 directly determine the output parameter dose, and thus have better correlations compared to the other predictor variables. The linear regression model with one variable (I-131 concentration in drinking water) was found to be the best with a root mean square error of 0.0133. For the other models, the root mean square errors are0.0148 for ridge regression cross validation,0.0198 for ridge regression L-curve, 0.0210 for PCR,0.0856 for PLS, 0.0892 for locally weighted linear regression, and 0.0993 for locally weighted kernel regression.

Index Terms—Nuclear accident, partial least square regression, principal component regression, radiation dose, radioactive materials, regression models, ridge regression.

Stephen U. Egarievwe is with Nuclear Engineering and Radiological Science Center, Alabama A&M University, Normal, AL 35762 USA (email: stephen.egarievwe@aamu.edu).
Jamie B. Coble, Laurence F. Miller are with Nuclear Engineering Department, University of Tennessee, Knoxville, TN 37996 USA.

Cite: Stephen U. Egarievwe, Jamie B. Coble, Laurence F. Miller, "Analysis of How Well Regression Models Predict Radiation Dose from the Fukushima Daiichi Nuclear Accident," International Journal of Applied Physics and Mathematics vol. 6, no. 4, pp. 150-164, 2016.

General Information

ISSN: 2010-362X
Frequency: Bimonthly (2011-2014); Quarterly (Since 2015)
DOI: 10.17706/IJAPM
Editor-in-Chief: Prof. Haydar Akca
Abstracting/ Indexing: Index Copernicus, EI (INSPEC, IET), Chemical Abstracts Services (CAS), Electronic Journals Library, Engineering & Technology Digital Library, Nanowerk Database, Google Scholar and ProQuest
E-mail: ijapm@iap.org
  • Mar 11, 2016 News!

    IJAPM Vol.5, No.3 has been indexed by EI (Inspec).   [Click]

  • Aug 10, 2017 News!

    The papers published in Vol 7, No 3 have all received dois from Crossref

  • Jul 21, 2017 News!

    Vol 7, No 3 has been published with online version     [Click]

  • Feb 22, 2017 News!

    The papers published in Vol.7, No.2 have all received dois from Crossref.

  • Feb 13, 2017 News!

    Vol.7, No.2 has been published with online version.   [Click]

  • Read more>>