Neural net classification of proton- and heavy nucleus-induced cosmic ray families

A feed-forward neural network trained using backpropagation is applied to discriminate between proton-induced cosmic ray families and heavy nucleus-induced ones. Fifteen input variables which characterize three-dimensional behaviour of the families are chosen. The network successfully classify the events with classification efficiency 85%. The trained neural network classifier is applied to the cosmic ray families observed in the Pamir chambers. The fraction of heavy nucleus-induced events is estimated, from the network-output distribution, to be at most 3% of the observed families. (author)