Comparison of different ML methods applied to the classification of events with ttbar in the final state at the ATLAS experiment

TL;DR

This paper compares various machine learning techniques for classifying top-antitop events in ATLAS data, focusing on their effectiveness and robustness in distinguishing BSM from SM events using simulated kinematic features.

Contribution

It provides a comprehensive comparison of ML methods applied to particle physics event classification, including hyper-parameter tuning and ensemble techniques.

Findings

Certain ML methods outperform others in classification accuracy.

Hyper-parameter optimization significantly impacts model performance.

The study offers insights into the robustness of different ML approaches.

Abstract

This contribution describes the experience with the application of different Machine Learning (ML) techniques to a physics analysis case. The use case chosen is the classification of top-antitop events coming from BSM or from SM using data from a repository of simulated events. The features of these events are represented by their kinematic observables. The initial objective was to compare different ML methods in order to see whether they can lead to an improvement in the classification, but the work has also helped us to test many variations in the methods by changing hyper-parameters, using different optimisers, ensembles, etc. With this information we have been able to conduct a comparative study that is useful for ensuring as complete control as possible of the methodology.