Surveys with the next generation of large radio telescopes are expected to yield thousands of new detections of HI 21-cm absorption at z > 0.1. This absorption traces the cold phase of the neutral ISM, the reservoir for star-formation, over the history of the Universe and can either be detected in cool gas associated with a radio source or from gas within a galaxy intervening the sight-line to the continuum source. Currently, there are only ~50 detections of each intervening and associated
absorption. The nature of the absorbing galaxy is usually determined from previous optical observations, which give the redshift of the continuum source, thus determining whether the absorption arises within an active galaxy (associated with the radio source) or a quiescent galaxy (intervening the radio source). With the large number of new detections expected, in conjunction with the fact that optical pre-selection may introduce a bias against the detection of 21-cm absorption, it is
therefore necessary to find a new technique which can determine the nature of these newly discovered gas-rich galaxies without the use of optical spectroscopy. We describe the application of machine learning techniques which, using the currently limited sample, exhibit a >80% accuracy in the prediction of absorber type.