Percentage of outliers (top) and non-reduced RMS error σ_{Δz/ (1 + z)} (bottom) in the photo-z estimation with the inclusion of multiple principal components of the morphological parameters with the AEGIS data set discussed in Sect. 6, starting with none, adding the first morphological principal component, then the first and the second principal components, and so on. Results are shown for SPIDERz (solid lines and star points) and for the custom neural network described in Singal et al. (2011, dashed lines and cross points). The uncertainties represent the standard deviation of the values obtained from different realizations featuring different compositions of the training and evaluation sets, as discussed in Sect. 6.2.