The composition of the cosmic rays in the knee region (≈ 104 TeV/nucleus) of all particle spectrum is considered to be the result of the particle acceleration and propagation from the astrophysical sources. The steeply falling cosmic ray spectrum makes a direct measurement of the composition difficult, but it can be inferred from the measurements of the showers generated by the interaction of the primary cosmic ray with the Earth atmosphere. In particular the characteristics of the muon bundles produced in the showers depend on the primary cosmic ray nature. The ANTARES telescope is situated 2.5 km under the Mediterranean Sea off the coast of Toulon, France. It is taking data in its complete configuration since 2008 with nearly 900 photomultipliers installed on 12 lines. The trigger rate is a few Hz dominated by atmospheric muons. A method using a multiple layered neural network as a classifier was developed to estimate the relative contribution of proton and iron showers from the energy and multiplicity distribution of the muon tracks reaching the detector. The performance of the method estimated from simulation will be discussed.
