Forecasting Pediatric Ward Census:A Machine Learning Model to Optimize Bed Allocation and Reduce Operational Waste

Abstract

Efficient bed allocation in pediatric wards is a persistent challenge for hospital operations due to strong seasonality, demand uncertainty, and limited resource flexibility. Inappropriate bed allocation may lead to congestion during peak periods and idle capacity during low-demand periods, resulting in operational waste and reduced service quality.
This study proposes a machine learning–assisted forecasting framework for pediatric ward census prediction using historical hospital data. A classical time-series model is employed as the baseline to capture long-term trends and seasonal patterns, while a machine learning–based residual learning strategy is introduced to adapt to short-term demand fluctuations. The forecasting results are further translated into actionable indicators for bed allocation and staff scheduling.
A case study based on real pediatric ward census data demonstrates that the proposed framework provides stable short-term forecasts and supports proactive hospital resource management. Rather than pursuing extreme prediction accuracy, the study emphasizes operational usability and decision relevance, offering a practical solution to reduce congestion risk and idle bed capacity in pediatric wards.