Classification of radioactive waste drums using random forests for their -ray spectra

Hata, Haruhi  ; Ishimori, Yuu

The feasibility of Random Forests, one of machine learning methods was examined for the classification of radioactive waste drums. It was carried out using 954 -ray spectra of drums which were already classified to natural or reprocessed uranium. After 300 spectra were selected at random to reassemble training datasets, the percentages of correct classification by Random Forests were evaluated with another 654 spectra. When the counts of spectra were reprocessed as the difference of their logarithm, Random Forests accurately classified 654 drums.