Rejection of Multi-jet Background in pp̄ → eν + jj̄ Channel through a SVM Classifier

Abstract

We test and optimize a multivariate discriminant software package, based on the Support Vector Machine (SVM) algorithm, to reduce the multi-jet background events in the channel pp̄ → eν + jj̄. We use the CDFII data-set, collected at the TeVatron pp̄ collider, where this channel provides the signature for many important physics processes: e.g. associated Higgs production, WZ, single top events. The Multi-jet background can be large and difficult to reject but, in this paper, we show that an appropriatly trained SVM can handle it in an effective way. The developed programs perform training set selection, efficiency maximization and consistency checks; we also discuss the robustness of the discriminant. A classification accuracy ≥ 95% can be reached using Monte Carlo simulated signal and a data-driven background model (limited by statistic) with a background rejection of ≃ 90%.