Improving ANAIS-112 sensitivity to DAMA/LIBRA signal with machine learning techniques

The DAMA/LIBRA observation of an annual modulation in the detection rate compatible with that expected for dark matter particles from galactic halo has accumulated evidence more than twenty years. It is only hint a direct elusive matter, but it strong tension negative results other very sensitive experiments, requiring ad-hoc scenarios to reconcile all present experimental results. Testing result using same target material, NaI(Tl), removes dependence on particle and models goal ANAIS-112 experiment, taking data at Canfranc Underground Laboratory Spain since August 2017 112.5 kg NaI(Tl). At low energies, dominated by non-bulk scintillation events careful event selection mandatory. This article summarizes efforts devoted better characterize filter this contribution boosted decision tree (BDT), trained high efficiency. We report training populations, procedure determine optimal cut BDT parameter, estimate efficiencies bulk region interest (ROI), evaluation performance analysis respect previous filtering. improvement achieved background rejection ROI, moreover, increase efficiency, push sensitivity test beyond 3$\sigma$ three-year exposure, being possible reach 5$\sigma$ extending few years scheduled 5 which were due 2022.