Multivariate selection of VLQ events

• Author: Zhaoyuan "Maxwell" Cui
• Advisor: Prof. Erich Varnes
• Physics department, University of Arizona

Introduction

This study leverages the TMVA tool kit of ROOT data analysis framework to enhance the search of vector-like quark (VLQ). The goal of the study is to use the multivariate analysis method (Machine Learning) to develop a new way of searhcing VLQ. Boosted Decision Tree (BDT) is used in this study. Background and signal raw data are generated by MC simulation and are provided by Sarah Jones.

Procedure

The procedure of the study contains two parts:

1. Data preparation
2. BDT training

Data preparation

The raw data from the simulation contain lots of infromation that is not needed in the training. At the same time, some information that one is interested, for example, b-jet, is not explicitly included. Therefore, it's necessary to extract and recombine the infromation from the raw data. Training data are generated by the code in the file prepTree.C. There are two specific modifications included (tick indicates done):

• b-tagging
• normalization of event weight

Variables that are included in training data file (tick indicates variables used as training input):

• jet_pt
• met_met
• met_sumet Scale sum of ET
• lep_pt
• ht Total visible transverse momentum
• bjet
• evtWeight

BDT Training

Since TMVA is a tool kit from ROOT, the language of TMVA is C/C++ and works like the usual ROOT operation.