Automatic selection of observables for the analysis of high-energy particle jets

A method is presented to reduce the size of a classification problem by automatically ranking the relative importance of available features. The variables are importance-sorted with a decision tree algorithm and correlated ones are removed after ranking. We tested the approach with the case of highly boosted di-jet resonances decaying to two b-quarks, to be selected against an overwhelming QCD background with a deep neural network (DNN). We make the relation between different importance rankings obtained with different algorithms explicit. We also show how the signal-to-background ratio changes, varying the number of features in input to the DNN.