QBDT, a new boosting decision tree method with systematic uncertainties into training for High Energy Physics

TL;DR

QBDT is a novel boosting decision tree method designed for high energy physics that directly optimizes significance and effectively incorporates systematic uncertainties, leading to improved signal detection and reduced uncertainty impact.

Contribution

This paper introduces QBDT, a new BDT approach that directly targets significance and systematically reduces the effect of uncertainties in high energy physics analyses.

Findings

QBDT reduces correlation between signal strength and uncertainties.

QBDT achieves 50-85% lower contribution to signal uncertainty compared to GradBDT.

QBDT improves significance in Higgs decay search.

Abstract

A new boosting decision tree (BDT) method, QBDT, is proposed for the classification problem in the field of high energy physics (HEP). In many HEP researches, great efforts are made to increase the signal significance with the presence of huge background and various systematical uncertainties. Why not develop a BDT method targeting the significance directly? Indeed, the significance plays a central role in this new method. It is used to split a node in building a tree and to be also the weight contributing to the BDT score. As the systematical uncertainties can be easily included in the significance calculation, this method is able to learn about reducing the effect of the systematical uncertainties via training. Taking the search of the rare radiative Higgs decay in proton-proton collisions $pp \to h + X \to \gamma\tau^+\tau^-+X$ as example, QBDT and the popular Gradient BDT (GradBDT) method are compared. QBDT is found to reduce the correlation between the signal strength and systematical uncertainty sources and thus to give a better significance. The contribution to the signal strength uncertainty from the systematical uncertainty sources using the new method is 50-85~\% of that using the GradBDT method.