A Neural Network Approach to High Energy Cosmic Rays Mass Identification

An event-by-event study, based on neural network methods, of the mass identification in high energy cosmic rays was carried out with simulated data, in order to check the possibility of analyzing real data measured at the Pierre Auger Observatory. Extensive air showers were simulated with the CORSIKA code, using the hadronic model QGSJET98. The goodness of the method in recognizing the mass of the primary was tested making use of the parameters extracted from the simulated longitudinal profiles. We showed that the designed supervised neuralnetwork is able to discriminate, with high identification efficiency and purity, between proton- and iron-induced showers. We tested our method also in presence of a four components primary flux (proton, helium, oxygen, and iron). Typical results for the classification matrix obtained are presented and discussed.