A conditional multi-label model to improve prediction of a rare outcome: An illustration predicting autism diagnosis

Wei A. Huang, Matthew Engelhard, Marika Coffman, Elliot D. Hill, Qin Weng, Abby Scheer, Gary Maslow, Ricardo Henao, Geraldine Dawson, Benjamin Alan Goldstein

Abstract: Highlights•We developed a conditional multi-label model to improve rare-event classification.•The conditional approach weights samples by the likelihood of any condition presence.•The multi-label approach captures shared features across autism and other conditions.•Combining both approaches yields the largest and most significant improvement.•Simulations show our model outperforms the baseline across various training sizes.