Forecasting emergency department occupancy with advanced machine learning models and multivariable input

• DOI: 10.1016/j.ijforecast.2023.12.002
• Publication Date: 2023-12-27T12:53:34Z
• Authors: Jalmari Tuominen; Eetu Pulkkinen; Jaakko Peltonen; Juho Kanniainen; Niku Oksala; Ari Palomäki; Antti Roine
• Keywords: Predictive modelling

Emergency department (ED) crowding is a significant threat to patient safety and it has been repeatedly associated with increased mortality. Forecasting future service demand the potential improve outcomes. Despite active research on subject, proposed forecasting models have become outdated, due quick influx of advanced machine learning because amount multivariable input data limited. In this study, we document performance set in ED occupancy 24 h ahead. We use electronic health record from large, combined an extensive explanatory variables, including availability beds catchment area hospitals, traffic local observation stations, weather more. show that DeepAR, N-BEATS, TFT, LightGBM all outperform traditional benchmarks, up 15% improvement. The inclusion variables enhances TFT DeepAR but fails significantly LightGBM. To best our knowledge, first study extensively superiority over statistical benchmarks context forecasting.