Machine Learning Applications in Neutrino Physics: Separation of Cherenkov and Scintillation Photons

In neutrino interactions, separation of Cherenkov and scintillation photons in a Water-based Liquid Scintillator (WbLS) is crucial for reconstructing neutrino energy, identification of secondary particles and distinguishing signal and background. In recent years, the use of artificial intelligence (AI) and machine learning (ML) methods has increased in the field of high energy physics, especially in neutrino physics. In this study, several machine learning-based classifier models were used to accurately separate of Cherenkov and scintillation photons and were compared with classical methods. This study focused on parameters such as the arrival time (hit-time) of photons resulting from neutrino interactions, their energies, and the 3D positions of photomultiplier tubes (PMTs) within the detector. Thus, more than 20 machine learning-based classification methods well-known in the literature were trained on two different balanced and unbalanced datasets. According to the results, the Random Forest, XGBoost, and Light GBM models have higher accuracy compared to other models. In addition, an ensemble model was developed by applying parameter optimization to these models. This model separates Cherenkov and scintillation photons with ~95% accuracy. Compared to traditional methods, machine learning based classifiers demonstrated significantly improved accuracy to separate Cherenkov and scintillation photons, and a comprehensive study has been conducted for models that should be focused on in the field of neutrino physics. In this study, we will comprehensively examine the machine learning-based classification models we tested in the field of neutrino physics and share the results of the parametric analyses applied to the dataset.