Identification of charged particles in a multilayer detector by the energy loss technique may also be achieved by the use of a neural network. The performance of the network becomes worse when a large fraction of information is missing, for instance due to detector inefficiencies. Algorithms which provide a way to impute missing information have been developed over the past years. Among the various approaches, we focused on normal mixtures models in comparison with standard mean imputation and multiple imputation methods. Further, to account for the intrinsic asymmetry of the energy loss data, we considered skew-normal mixture models and provided a closed form implementation in the Expectation-Maximization (EM) algorithm framework to handle missing patterns. The method has been applied to a test case where the energy losses of pions, kaons and protons in a six-layers Silicon detector are considered as input neurons to a neural network. Results are given in terms of reconstruction efficiency and purity of the various species in different momentum bins.