Separating γ- and hadron-induced cosmic ray air showers with feed-forward neural networks using the charged particle information⋆

Abstract The success of current and future air shower arrays in detecting point sources of cosmic rays above 10 TeV depends crucially on the possibility of finding efficient methods for separating γ -induced air showers from the overwhelming background of hadron-induced showers. We study the application of computer-simulated feed-forward neural networks in the analysis of cosmic ray data taken with the Geiger towers of the HEGRA air shower array. The combination of these charged particle detectors with the neural net based analysis is characterized by high background rejection and signal efficiency. In contrast to the often-quoted non-transparency of the net technique, the detailed analysis of the net performance gives insight into the physics involved and helps to asses the different information that allow γ /hadron separation