Topics in statistical data analysis for high-energy physics

TL;DR

This paper discusses advanced statistical techniques in high-energy physics, focusing on Bayesian methods and multivariate analysis, including boosted decision trees, to improve data interpretation and event classification.

Contribution

It introduces the application of Bayesian statistics and multivariate methods, especially boosted decision trees, to enhance data analysis in high-energy physics.

Findings

Bayesian approach allows direct probability assessment of hypotheses.

Multivariate methods improve event classification accuracy.

Boosted decision trees are effective in distinguishing event types.

Abstract

These lectures concern two topics that are becoming increasingly important in the analysis of High Energy Physics (HEP) data: Bayesian statistics and multivariate methods. In the Bayesian approach we extend the interpretation of probability to cover not only the frequency of repeatable outcomes but also to include a degree of belief. In this way we are able to associate probability with a hypothesis and thus to answer directly questions that cannot be addressed easily with traditional frequentist methods. In multivariate analysis we try to exploit as much information as possible from the characteristics that we measure for each event to distinguish between event types. In particular we will look at a method that has gained popularity in HEP in recent years: the boosted decision tree (BDT).