The success of most revenue management systems is based solely on demand forecasting systems. The better such a system predicts demand, the easier it is to manage limited resources and prices, which leads to better utilization and higher revenue. In this master's thesis, we propose two different demand forecasting approaches, one of which consists of one model and the other of several regression models based on booking windows. We compare the proposed method with the persistence model, the statistical ARIMA model and a similar model from a related work. In the thesis we also conduct an experiment with a domain expert and compare his results with our approach. We test the methods on real data of a specific hotel, for which we also develop an application programming interface for obtaining demand forecasts. The results show that the proposed methods are the best among those considered, as the forecast errors are smaller than the errors of the other methods and the domain expert's errors.
|
