University of Houston

Problems in High-Energy Particle Physics commonly involve sifting large amounts of data in search of a specific pattern of behavior. Due to size and accuracy requirements, this process is intractable if handled manually, but it is an excellent candidate for binary classification techniques using machine learning. We consider the nature and challenge of applying such techniques and implement three specific algorithms within the Root framework developed at the European Organization for Nuclear Research. Our primary approach is Bayesian, using mixtures of Cauchy and Gaussian distributions to model the complex feature space, and Expectation Maximization to derive the model parameters. Since performance is a concern with large datasets, we implement an enhanced version of Gaussian mixture models with significantly better convergence speed, and improve this approach further to preserve accuracy.