Short-term forecasting of air cargo demand from a European airport hub to the United States during COVID-19

Air cargo is mostly transported on passenger flights. During the COVID-19 outbreak, there have been worldwide restrictions on passenger transportation. Therefore, airlines experienced a capacity problem for air cargo. Better insight of air cargo demand during COVID-19 could contribute to the better arrangement of capacity by accordingly adapting flight schedules for cargo. The aim of this research was to make short-term predictions of air cargo demand between a major European airport hub and the United States during the COVID-19 pandemic. This was done for the month of May in 2020 by making 14-day predictions. The same was done for the year 2019 to observe whether the models performed well in the absence of the pandemic. The data set was compiled using data provided by a major commercial airline and exogenous features, such as stock market indices, foreign currency exchange rates and healthcare related predictions during COVID-19. To make the predictions, two classes of machine learning models for time series were compared: Autoregressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM). In the year 2020, the best performing model among the ARIMA-based models is the Seasonal ARIMA including the exogenous feature Schedule . During the year 2019 the Seasonal ARIMA model without exogenous features generates the most accurate predictions. Among the LSTM models, the multivariate LSTM models outperform the univariate LSTM models in both years. Nonetheless, the ARIMA-based models are more accurate than the multivariate LSTM model in this research.