Time-series forecasting: Air GDS bookings prediction

During my internship as a data analyst at Amadeus, I developed a forecasting project for Air GDS bookings time-series. GDS (Global Distribution System) is a digital system enabling flight tickets bookings for travel agencies. The main aspects of the project are the high number of time-series to analyse and forecast and the huge impact that Covid pandemic had on data. After analysing data, I applied two different models, one requiring more expertise in time-series modeling, and the second one accessible also to analysts with less mathematical and statistical background: SARIMAX and Prophet. SARIMAX is a statistic regression model with residuals modeled as Seasonal ARIMA time-series. I studied a different model for each time-series, validating them both with statistical tests and with the error on the validation data set using a three months horizon forecast. Prophet is a time-series forecasting model developed by Facebook data science team. It is a generalized additive model containing interpretable parameters that can be intuitively adjusted. Also in this second method, I modeled each time-series individually, tuning the parameters on the same validation data set. Comparing the methods on a test set, SARIMAX turned out to provide better results than Prophet: this is the reason why I chose the former for generating the final forecast for Air GDS Bookings. The language used to analyze and model the data was R, while Qlik Sense software was used to build the output visible to the business users.