A K-means Group Division and LSTM Based Method for Hotel Demand Forecasting

The accuracy of hotel demand forecasting is affected by factors such as the completeness of historical data and the maturity of models. Most of the existing methods are based on rich data, without considering that single hotels may only obtain sparse data. Therefore, a K-means group division and Long Short-Term Memory (LSTM) based method is proposed in this paper. Guest types are introduced into the forecasting to provide reference for hotel's further decision-making. Using an example of 1493 hotels in Europe, we divide hotel groups and forecast the flow of leisure and business guests. The experimental results show that, compared with the benchmark models, LSTM can improve the forecasting performance of hotel group; compared with single hotels, the forecasting of hotel groups can effectively avoid inaccuracy caused by sparse data. The results can provide necessary reference for hospitality to make decisions based on guest types.