[en] Full text: The study of relativistic heavy ion interactions is motivated by its potential to reveal the creation of a new state of hadronic matter, the Quark Gluon Plasma (QGP). The existence of QGP has been postulated to occur at very high energy densities, in excess of 3GeV.fm^-3. Such energy densities are believed to be achievable in very central, head-on collisions of heavy ions. For this reason, the concept of the 'centrality' of a collision is crucially important for the pre-selection, classification and analysis of event data. However, since the impact parameter of a collision is not directly measurable, its accurate estimation has to rely on other experimentally observable event variables and features. The complexity and the fluctuations of this process can make this a very difficult task. In this work we investigate the capabilities and limitations of impact parameter estimation by means of various neural network architectures and algorithms. The performance of the neural network classifiers is compared with traditional methods of impact parameter estimation