Combining Process Mining and Time Series Forecasting to Predict Hospital Bed Occupancy

Abstract

This research investigates in how far AI methods can support the prediction of bed occupancy in hospital units based on individual patient data. We combine process mining and a Deep Spatial-Temporal Graph Modeling algorithm and show that this improves the performance of the prediction over existing approaches. To improve the model even more it is extended with knowledge available from patient records, like the day of the week, the time of the day, whether it is a vacation day or not and the amount of emergency cases per data point.