Machine learning models to determine unobservable centrality-related parameter values for a wide range of nuclear systems at the energy of 200 GeV

Abstract

In the paper, a comparative analysis and a search for the optimal machine learning model have been conducted. The model should predict the values of unobservable centrality-related quantities based on the experimental data for observable quantities, namely, the number of charged particles and the number of neutral ones born in the interactions of both heavy and light ultrarelativistic nuclei. The sought-for unobservable values were the numbers of wounded nucleons involved in the interactions and of the binary nucleon-nucleon collisions. Linear and polynomial regressions of various degrees, a decision tree (DT), a random forest (RF), and a multilayer perceptron (MP) were chosen and considered as machine learning models. The prediction accuracy of the models was characterized and tested by the coefficient of determination. The DT, RF, and MP models were found to predict the desired values with the highest accuracy, i.e., they gave equally good results.

Keywords

• machine learning
• nuclei collisions
• regression
• decision tree
• random forest
• multilayer perceptron