Accuracy Correlation in Neutron Resonance Reclassification

Collecting accurate neutron resonances is essential for application in practical nuclear systems and understanding astrophysical processes. Current methods for finding the resonance quantum numbers associated with angular momenta and spin are subjective and irreproducible, often leading to incorrect spin assignments. To solve this problem, we have employed a machine learning (ML) method to train an algorithm for identifying and reclassifying incorrect neutron spin assignments. Currently, the algorithm operates with varied successes depending on the isotope. For this project, we are examining the properties of the algorithm on polarized In-115. We build synthetic data that mimics the statistical properties of real resonances to train the algorithm. We then validate the trained algorithm with a set of real In-115 data and observe the correlation between the two sets. However, for unpolarized data, we cannot guarantee the given resonances as accurate, so we also test the trained algorithm on an In-115 set with jumbled resonance assignments. We can then improve the validation accuracy by adjusting the ML classifier's parameters. We also explored an iterative method in which successive reclassifications could incrementally improve the quality of any misclassified resonance sequence.