Multivariate analysis of Compton-imaging detector response for neutron capture cross-section measurements

Abstract: The imaging total-energy detector (i-TED) uses Compton imaging to suppress background in radiative neutron capture measurements, but stringent geometric selection often entails substantial losses in event statistics. This work investigates a TabM-based multivariate event-selection method using measurements at the back-streaming white neutron beamline (Back-n) of the China Spallation Neutron Source (CSNS). Capture-cascade, particle-transport and detector-response simulations are combined with $^{197}$Au sample measurements to evaluate background suppression and event retention. Experimental results show that TabM retains more coincidences than geometric selection at the same effect-to-background ratio. At equal selected counts, it achieves a higher ratio. Pulse-height weighting and selection-specific response calibration are combined with the corresponding experimental corrections to reconstruct the $^{197}$Au(n,$\gamma$) cross section. Both reconstructions retain the main Au resonance structures, and uncertainty propagation shows improved precision for TabM with response calibration and selection thresholds held fixed. In a separate weak-signal $^{209}$Bi simulation, multivariate selection retains more true capture events at equal selected counts, making resonance structures that are weak under geometric selection more distinct. The method improves the statistical efficiency of i-TED capture measurements and provides a basis for measurements of weak capture signals and mass-limited samples.