Skip to Main Content
The problem of identifying cosmic gamma ray events out of charged cosmic ray background in Cherenkov telescopes is one of the key problems in very high energy gamma ray astronomy. Separation between gamma-like and hadron-like events is performed by a Bayesian ensemble of neural networks and Markov chain Monte Carlo methods for model parameters optimization. The results are discussed in terms of the energy of the primaries and a complete study is made by using various data representation methods with different levels of feature reduction. Our classifier clearly outperforms the results obtained using standard feedforward neural networks, and its performance is comparable with random forests, which is actually used in data analysis of the MAGIC Cherenkov telescope. Regarding the energy of the primaries, it achieves very promising results in terms of classification accuracy with low energy events, the most difficult and unexplored energy range which will be a major issue in future explorations.