Deep neural network predictions for excitation functions of ¹⁶⁵Ho(α,xn) reactions

Accurate nuclear reaction cross section data are essential for nuclear medicine, reactor technology, and nuclear astrophysics. In this study, excitation functions for ¹⁶⁵Ho(α,n)¹⁶⁸Tm, ¹⁶⁵Ho(α,2n)¹⁶⁷Tm, ¹⁶⁵Ho(α,3n)¹⁶⁶Tm and ¹⁶⁵Ho(α,4n)¹⁶⁵Tm reactions are analyzed over a wide range of alpha incident energies. Experimental data from EXFOR are compared with theoretical predictions generated using the TALYS nuclear reaction code and the TENDL-2023 evaluated nuclear data library. Additionally, a data-driven approach utilizing Deep Neural Networks (DNNs) with various activation functions (ReLU, ELU, LeakyReLU, SiLU, Mish, PReLU) is developed to predict the cross sections. Python programming language and pytorch module are used in the DNN predictions. The results demonstrate that while conventional models provide a reasonable approximation of reaction trends, Artificial Neural Network (ANN) models which are a branch of machine learning significantly improve agreement with experimental data. These findings underscore the potential of artificial intelligence as a complementary tool for enhancing nuclear reaction modeling. In addition, using different activation functions in the deep learning algorithm is important to get the best results in the predictions of the experimental data.