Multivariate statistical methods and data mining in HEP
Course description: The lectures will cover multivariate statistical methods and their applications in High Energy Physics. The methods will be viewed in the framework of a statistical test, as used e.g. to discriminate between signal and background events. Topics will include an introduction to the relevant statistical formalism, linear test variables, neural networks, probability density estimation (PDE) methods, kernel-based PDE, decision trees and support vector machines. The methods will be evaluated with respect to criteria relevant to HEP analyses such as statistical power, ease of computation and sensitivity to systematic effects. Simple computer examples that can be extended to more complex analyses will be presented.
The lectures will be in the CERN main auditorium from Monday 16 June to Thursday 19 June, 2008 at 11:00.
Lecture Notes in pdf format (preliminary -- subject to minor changes):
The code used to make some of the simple examples with the TMVA package can be found here.