Improving the NOvA 3-Flavour Neutrino Oscillation Analysis Event Selection with a Neural Network Based Selection Algorithm

NOvA is a long-baseline neutrino oscillation experiment, consisting of two functionally identical tracking calorimeter detectors deployed in the Fermilab NuMI beam. Both NOvA detectors are placed 14.6 mrad off the beam axis to achieve a narrow-band neutrino flux at about 2 GeV, which is an oscillation maximum at the far detector placed 810km from the target. NOvA has two main oscillation channels: νμ disappearance and νe appearance as well as the anti-neutrino equivalents.

The NOvA 3-flavour neutrino oscillation analysis has a good and robust event selection, which ensures that background particle interactions are minimized in the analysis. The current νμ selection has a very high efficiency, but more events could be recovered to improve the sensitivity and statistics of the results. This talk presents a neural network selection algorithm that was developed to recover more of these events back into the analysis. The network was trained to divide the data into signal (νμ CC events) and background (NC and νe CC events). Initial results show that these events can bring improvement to the sensitivity to neutrino oscillation parameters, and a cut based on this network is currently being developed.