(A) study on improved method for scaling factor calculation with robust regression

Abstract

The radioactivities of specific nuclides in waste packages have to be declared in accordance with limits and criteria derived from safety assessment of the disposal facility. Some of these nuclides are difficult to measure from the outside of the waste packages because they are alpha or beta emitting nuclides. There are a number of activity determination methods. The scaling-factor method is widely applied in order to evaluate these difficult-to-measure nuclides. The scaling-factor method is based on a correlation between easily measurable gamma emitting nuclides and difficult-to-measure nuclides. Scaling-factor has been calculated by the statistical processing of radiochemical analysis data. Current statistical methods to calculate scaling-factor are geometric mean and the linear regression based on least square. However both methods have a chance to distort the scaling-factor because they are vulnerable to outliers especially in small number of data points. In this study, robust regression (Least Median square, Least Trimmed Square) is suggested and applied to calculate scaling-factor. These results are compared with the scaling-factors calculated by current statistical methods. Reweighted Least Square method that has the merits of both robust regression and current methods also suggested and verified. It is concluded that RLS method is effective to reduce the distortion of scaling-factor. It was shown that this method is applicable to some nuclides ($^3H$, $^{14}C$, $^{129}I$, $^{99}Tc$ and Gross-alpha) which they have low correlation coefficient caused by outliers.