Multiple linear regression analysis of scintillation gamma-ray spectra: Automatic candidate selection

Abstract Multiple linear regression (MLR) has been shown to be a quantitative and statiscally justifiable method for unfolding scintillation gamma-ray pulse-height spectra of radionuclide mixtures. The principal requirements of the method are:(1) that well-calibrated data are used, and (2) that candidate set of spectra (standard spectra to be used as regression variables) must be complete and exclusive. In the usual application of the MLR method, the user exercises a great deal of subjective (irreproducible) judgment in the selection of a candidate set. Several algorithms have been proposed for automatic candidate selection but are not completely justifiable on the basis of proper statistical procedures. In this report, an algorithm is described in which justifiable statistical procedures are used. Correlation coefficients are examined to determine the proper candidate set for a given experimental spectrum. The inclusion or exclusion of candidate spectra are justified on the basis of valid statistical indicators, and an estimate of the goodness of the final fit is provided. The results of unfolding actual experimental spectra by this method are included in the paper.