Radionuclide identification analysis using machine learning and GEANT4 simulation

Abstract

Security of nation border is an important thing which has to be kept from many kinds of criminal activities. One of the important things is preventing the nation from nuclear criminal activity like illegal nuclear source trafficking, nuclear source abuse and or nuclear pollution which want to enter to our country. Indonesia, through national nuclear agency (BATAN) will build environment and nuclear monitoring device station which will be placed at each state border. This device prepared to detect nuclear pollution around the station which potentially can harming the nation environment. Detecting nuclear pollution in the open environment is required to have an ability to radionuclide identifications rapidly. Support Vector Machine and Linear Discriminant Analysis was used in this research which prepared to be used on environment and nuclear monitoring device. Pre-processing was done to make the gamma spectrum data from CsI(Na) detector as feature can be analyzed with SVM and LDA. The setup of data collection was done with variation of source-detector distance, there are 50, 75, 100, 125 and 150 cm, variation of measurement time, from 1, 5, 10, 15, 30, 60 and 120 minutes. The result shows that LDA method can detect the radionuclide source with an accuracy value 84 percent, whereas using SVM-LDA method can get accuracy score more than 91 percent.