Optimization of Selection Criteria for B→D^(*)ηl⁺ν_l

The Belle experiment and its successor, Belle II, study B-mesons produced via e⁺e^- annihilation near the Υ(4S) resonance. Precise measurements of B-meson decays are important for our understanding of flavor-mixing and may yield insight into physics beyond the standard model. An interesting anomaly is the significant discrepancy between inclusive and exclusive measurements of the B→X_clν branching fractions. One possible decay mode that could bridge this gap is B→D^(*)ηl⁺ν_l. In this work, we use a boosted decision tree (BDT) to select the products of this semi-leptonic decay. We first seek to distinguish signal from background using variables that describe η and its subsequent decay to γγ. We determined discriminating input variables using simulated signal and background samples. Furthermore, BDT hyperparameters were tuned to minimize overtraining and maximize background rejection as a function of signal efficiency. This method notably improved the signal purity compared to a cut-based approach.