Forecasting of hospital bed occupation is a very important part of the hospital management. The usage of hospital resources should be optimized to minimize costs and maximize the quality of patient care. Forecasting of bed occupation enables tactical and strategical planning of the hospital logistic, in particular scheduling procedures or vacation periods for the personnel.

The goal of the project is to develop a machine learning algorithm for hospital bed forecasting in the scale of several months. The data is received from a medium size German hospital (ca. 500 beds) and covers approximately 12 years of admission/release information. Our solution is based on recurrent neural networks and shows promising performance. Openly available web-based implementation of the algorithm is in development.