Towards an Improved Estimator of the W Boson Recoil at High Pileup

This report describes a method to estimate the hadronic recoil of the W boson from high pileup events of W decay to a lepton and neutrino, using machine learning techniques (specifically boosted decision trees). Monte Carlo (MC) simulations of the decay have their limitations and a correction to data from the CMS (Compact Muon Solenoid) experiment is needed. Data of decays of Z bosons to two leptons were used to test that the method was able to estimate the hadronic recoil. Two different techniques one involving quantile regression and another involving vertical morphing are presented and compared. Quantile regression worked well for a small number of features in the training; looking at the difference between the corrected MC data and the detector data gave rise to a χ2 of 70.9. Whereas, quantile morphing worked well for a larger number of features; the difference here between corrected MC and detector data was found to create a χ2 of 167.