Comparison of classification performances between Random forests and Support vector machine for -ray spectral data of uranium-bearing waste drums

Hata, Haruhi   ; Ishimori, Yuu  

For the estimation of radioactive inventory, the radioactive waste drums should be classified based on their radioactive composition. We compared the classification performances between random forests and support vector machine, both of which are machine learning methods. The tested uranium in waste drums included natural uranium from uranium ore, reprocessed uranium from nuclear fuel, and natural uranium with rich radium from the impurities in yellow cake. A total of 75 data in 1037 -ray spectral data of these drums were trained, and 962 data were applied in the classification models. It was found that the random forests were advantageous in the shift of the channels.