We present the development of neutron-tagging techniques in Super-Kamiokande IV using a neural network analysis. The detection efficiency of neutron capture on hydrogen is estimated to be 26%, with a mis-tag rate of 0.016 per neutrino event. The uncertainty of the tagging efficiency is estimated to be 9.0%. Measurement of the tagging efficiency with data from an Americium-Beryllium calibration agrees with this value within 10%. The tagging procedure was performed on 3,244.4 days of SK-IV atmospheric neutrino data, identifying 18,091 neutrons in 26,473 neutrino events. The fitted neutron capture lifetime was measured as 218 \pm 9 \mu s