Model based clustering method as a new multivariate technique in high energy physics
Michal Štěpánek, Jiří Franc, V. Kůs · Journal of Physics Conference Series · 2014
Analysis of experimental data has one of the most important roles in High Energy Physics. Commonly used multivariate techniques such as Boosted Decision Trees or Bayesian Neural Networks are based on learning algorithms using Monte Carlo generated samples. We implemented a new Model Based Clustering method using Bayesian statistics and a modified iterative Expectation-Maximization algorithm for weighted data that have never been applied in this area. This greatly promising method was developed especially for the data collected from the DO experiment, which was one of two large particle physics experiments at the Tevatron proton-antiproton collider at Fermilab. We optimized and tested the proposed method in the single top search using a data sample of 9.7 fb−1 of integrated luminosity, which corresponds to the entire Run II DO dataset.