Forecast all at once: A multi-series forecasting approach for hospital discharges and admissions
Date: September 10 2026 10:45-11:30
Event: The Operational Research Society
Location: University of Nottingham, UK
Context
At the Operational Research Society Annual Conference (ORS 2026), I presented our research on “Forecast all at once: A multi-series forecasting approach for hospital discharges and admissions.” This presentation introduced a unified global multi-series forecasting framework designed to simultaneously predict daily admissions and discharges across 29 clinical specialties (58 interdependent time series). By capturing system-wide ripple effects through pooled cross-lag features and recursive multi-step forecasting with LightGBM, this approach replaces 58 individually tuned univariate models with a single scalable model, substantially reducing maintenance overhead while achieving superior point and probabilistic forecast accuracy to support hospital bed management and capacity planning.