Machine learning models to find unobservable centrality-related parameter values in collisions of different nuclei in the initial energy range from 40 to 200 GeV

Abstract

This paper continues studies in machine learning models capabilities aimed to finding the best way to predict the values of unobservable quantities that characterize centrality, based on experimental data for observable quantities: the number of charged particles and the number of neutrons produced in ultrarelativistic nuclear interactions. The sought-for unobservable quantities were the number of wounded nucleons involved in the interaction and the number of binary nucleon-nucleon collisions. A decision tree, a random forest, and a multilayer perceptron (MP) were tested as machine learning models. The prediction accuracy of the models was characterized by the coefficient of determination R2. Dependences of R2 values on initial energies (40 – 200 GeV) for different systems of colliding nuclei were obtained. The MP model was found to be able to predict the values of unknown quantities in a wide range of initial energies for different systems of nuclear interactions with good accuracy.

Keywords

• machine learning
• nuclei collisions
• initial energy
• r-squared
• multilayer perceptron