Detecting explosives by PGNAA using KNN Regressors and decision tree classifier: A proof of concept

Abstract Radiation based techniques such as PGNAA provided a good alternative to conventional explosives detection methods due to the simplicity and efficiency of the quantitative isotopic technique. This paper introduces a framework that identifies explosive materials using H, C, and O isotopic prints. The first step is to regress the weight fractions of H, C and O isotopes in the sample separately using the gamma peaks as input. The regressed percentages will be used as input to the classifier to identify if this sample is explosive or not. The data used for training the model are generated using MCNP5 and validated on the 2017 ROMASHA experimental setup in Frank Laboratory in JINR, Russia. Our data set consisted of 316 gamma peak observations, which are split into 85% for training and 15% for testing. Experiments showed that KNN-regressor achieved best results to predict the H, C and O weight fractions with average MSE of 0.005 and R 2 of 0.95. Also, the decision-tree-classifier achieved best results to identify whether the sample is explosive or not with the accuracy of 0.98. And, the assembled pipeline achieved total accuracy of 0.92 after error propagation through the two models. The proposed framework emphasized that machine learning and PGNAA are capable of learning and identifying explosives with the accuracy of 92%.