Simplex optimization of artificial neural networks for the prediction of minimum detectable activity in gamma-ray spectrometry

Abstract

A three-layer feed-forward artificial neural network (ANN) with a back-propagation learning algorithm was used to predict the minimum detectable activity (AD) of radionuclides (Ra-226, U-238, U-235, K-40, Th-232, Cs-134, Cs-137 and Be-7) in environmental soil samples as a function of measurement time. The ANN parameters (learning rate, momentum, number of epochs, and the number of nodes in the hidden layer) were optimized simultaneously employing a variable-size simplex method. The optimized ANN model revealed satisfactory predictions, with correlation coefficients between experimental and predicted values 0.9517 for 232 Th (sample with U-238/Th-232 ratio of 1.14) to 0.9995 for K-40 (sample with U-238/Th-232 ratio of 0.43). Neither the differences between the measured and the predicted A(D) values nor the correlation coefficients were influenced by the absolute values of AD for the investigated radionuclides.