Implementation of sampling techniques for initial geometry prediction in heavy ion collision experiment

We show that three crucial features that determine the initial geometry of heavy ion collision (HIC) experiments can be predicted with excellent accuracy by employing supervised machine learning (ML) techniques. Despite the use of several ML techniques in the past, the prediction accuracy is hugely centrality dependent. Detailed parameter scans and ablation analyses are used to analyse the error spectrum. We use different sampling methods to determine an efficient algorithm that provides a multi-fold improvement in the accuracy of ML model prediction. We discuss how the errors can be minimized, and the accuracy can be improved to a great extent in all the ranges of impact parameter and eccentricity predictions