Development of an Automated Pipeline for Forecasting Time Series¶of the Number of Bookings

Modern business needs to be introduced to the data-driven culture, which implies making decisions based on regularly-parsed data. In the case of the hotel business, the optimal solution is to develop an automated pipeline that stores, processes and analyses regularly data every day. For such a task the time series analysis tools (ARIMA, ARIMAX models), tools for operations with large number of features and decompositions (Prophet, Random Forest, LightGBM), as well as development of all backend part (PostgreSQL database and DBeaver DBMS) and its automation (Cron) were required. The demand sensitivity model (IRF) was also trained on the Price parameters of the average tariff and traffic volume on the company's website, and a cross- validation was carried out for different timeframes. As a result, LightGBM model was selected to apply to each hotel pair and room type, which together showed the best predictive accuracy on the cross-validation data. These models are automatically placed in the configured databases, updated daily, and have a view that is ready to be used by business units.