Forecasting Hospital Resource Demand Using Gradient Boosting: An Operational Analytics Approach for Bed Allocation and Patient Flow Management

Abstract

Hospital resource management faces increasing complexity due to volatile patient demand and capacity constraints. This research presents a hybrid forecasting framework integrating time series decomposition with gradient boosting techniques for predicting hospital bed occupancy and patient flow patterns. Using three years of operational data from a large American hospital system, the proposed approach combines seasonal decomposition methods with LightGBM to capture both temporal patterns and complex non-linear relationships. Experimental results demonstrate mean absolute percentage error of 2.3% for one-day-ahead bed occupancy predictions, representing 18% improvement over standalone machine learning methods and 32% improvement over classical time series approaches. The framework successfully forecasts emergency department volumes with 6.4% mean absolute percentage error while maintaining computational efficiency suitable for daily operational deployment. Implementation case studies reveal measurable operational improvements including 17% reduction in bed assignment times and enhanced equipment utilization. This research contributes a practical methodology for transforming reactive hospital resource management into proactive capacity planning.