Effects of linear mass decorrelation on W boson tagging with multivariate classifiers
Abstract
High energy physics multivariate machine learning classifiers that distinguish W boson jets from QCD interaction jets have input variables that are often highly correlated with invariant jet mass; this results in a multivariate classifier output highly correlated with invariant jet mass as well. This thesis investigates the effect of an algorithm to remove all linear correlation with QCD jet mass from input variables of the multivariate classifier. The results of this comparison show removing linear correlation with mass from the input variables is not sufficient to remove the correlation between classifier output and invariant jet mass.