By means of Monte Carlo methods, we perform a full error analysis on the Duflo-Zucker mass model. In particular, we study the presence of correlations in the residuals to obtain a more realistic estimate of the error bars on the predicted binding energies. To further reduce the discrepancies between model prediction and experimental data we also apply a Multilayer Perceptron Neural Network. We show that the root mean square of the model further reduces of roughly 40\%. We then use the resulting models to predict the composition of the outer crust of a non accreting neutron star. We provide a first estimate of the impact of error propagation on the resulting equation of state of the system.