Particle identification in multi-detector arrays using machine learning

Particle identification (PID) is a crucial component to a wide variety of nuclear physics analyses and is

done in a number of ways including pulse shape discrimination and the ΔE-E technique. These

applications tend to rely on a tedious `by-hand’ selection of clusters in the data. In large multi-detector

arrays, this process can sometimes take years to perform. A new method of selecting the clusters of

data in the PID-space using unsupervised machine learning techniques is proposed to automate and

vastly accelerate the particle identification process. We apply a modern clustering algorithm called

hierarchical density based spatial clustering for applications with noise (HDBSCAN) augmented with

principle-component analysis. This method shows promise to perform elemental and isotopic

identification in ΔE vs E. The ability to identify particles in ΔE-E-space - or other PID spaces - with an

automated method could even permit preliminary physics analysis in large multi-detector arrays even as

experimental data is coming in.