The star-forming reservoir in the distant Universe can be detected through HI 21-cm absorption arising from either cool gas associated with a radio source or from within a galaxy intervening the sightline to the continuum source. In order to test whether the nature of the absorber can be predicted from the profile shape, we have compiled and analysed all of the known redshifted (z>=0.1) HI 21-cm absorption profiles. Although between individual spectra there is too much variation to assign a typical spectral profile, we confirm that associated absorption profiles are, on average, wider than their intervening counterparts. It is widely hypothesized that this is due to high-velocity nuclear gas feeding the central engine, absent in the more quiescent intervening absorbers. Modelling the column density distribution of the mean associated and intervening spectra, we confirm that the additional low optical depth, wide dispersion component, typical of associated absorbers, arises from gas within the inner parsec. With regard to the potential of predicting the absorber type in the absence of optical spectroscopy, we have implemented machine learning techniques to the 55 associated and 43 intervening spectra, with each of the tested models giving a >=80 per cent accuracy in the prediction of the absorber type. Given the impracticability of follow-up optical spectroscopy of the large number of 21-cm detections expected from the next generation of large radio telescopes, this could provide a powerful new technique with which to determine the nature of the absorbing galaxy.