Optimization Studies for the H $\rightarrow$ WW Boosted Decision Tree Analysis

The aim of this project was to follow the ATLAS $H \rightarrow WW$ BDT analysis and try to optimize training variables, pre-selection cuts, and training parameters such as the depth \cite{orig, spin}. Machine learning was done with Monte Carlo samples of the $H \rightarrow W^+ W^- \rightarrow e \mu \nu _e \nu_\mu$ channel. Multivariate analysis (TMVA) was executed by way of boosted decision tree (BDT) in attempt to improve the original ATLAS $H \rightarrow WW$ BDT analysis. The goal of the machine is to separate the Higgs signal from the continuous WW background only. Once an optimal set-up was found and used for training, the weights produced from the BDT output were used categorize an unknown data set. Finally, regional cuts were made to the final BDT output to observe the performance of the training, and it appeared to perform well.