Deep Learning-Based Forecasting of Boarding Patient Counts to Address ED Overcrowding

Emergency department (ED) overcrowding remains a major challenge for hospitals, resulting in worse outcomes, longer waits, higher costs, and greater strain on staff. Boarding count—the number of admitted patients waiting for a bed—is a key patient flow metric that affects overall ED operations. This study presents a deep learning-based approach to forecast ED boarding counts six hours in advance using only operational and contextual features, without patient-level clinical data. Hourly data from ED tracking, inpatient census, weather, holidays, and local events were merged and preprocessed for model training. Multiple time series deep learning models were evaluated, including the CNN-based ResNetPlus and two transformer-based models, TSTPlus and TSiTPlus, with hyperparameters optimized using Optuna. TSTPlus achieved the best performance, with a mean absolute error of 4.30, mean squared error of 29.47, root mean squared error of 5.43, and an R² score of 0.79. Model explainability was analyzed to interpret the influence of different input features on predictions. The models were also tested under extreme scenarios, demonstrating accurate forecasts even in challenging conditions. Results show the models are effective for six-hour-ahead boarding count prediction, aiding proactive planning to reduce ED overcrowding. The source code and model implementation are available at https://github.com/drorhunvural/ED_OverCrowding_Predictions.