The 6th International Conference on Research in Air Transportation (ICRAT) was held the week of May 26-30, 2014 at the Istanbul Technical University. The ICRAT has now been established as a mainstream biennial event in Air Transport Research, alternating with the USA/Europe Air Traffic Management (ATM) Research and Development (R&D) Seminar. In this 2014 edition almost two hundred researchers and air transportation stakeholders attended ICRAT, which included several ComplexWorld activities: tutorials, key note speeches, and paper sessions on complexity & ATM. Amoung them, we would like to highlight the remarkable interest that piqued in the tutorial session brilliantly provided by Massimiliano Zanin (Innaxis) and Samuel Cristóbal (Innaxis), on different areas within Data Science in air transportation. In fact, it was the parallel session with the highest number of attendees, congrats!
The aviation sector gathers and stores a large amount of unstructured, heterogeneous data from different sources and of diverse natures: safety data and reports, flight plans, navigation data, weather, airport data, radar tracks, etc. From airlines to ANSPs or airports, the ability to collect information from different data sensors is growing exponentially. Nevertheless, how the different stakeholders take advantage of this data has not evolved as rapidly and there is still much room for improvement. In this talk, Massimiliano reviewed the topic of utmost importance for the correct application of data analysis to air transport: stationarity.
Stationarity is the property of a system having coherent characteristics in time and space, the latter being both the physical space (i.e. the position of airports throughout the world) and the virtual space created by the parameters of the system. When stationarity cannot be guaranteed, the results obtained can be plagued with errors and inconsistencies. For instance, when analysing the time series representing some observables, causal relations may appear: yet they may just be the result of some constant trend, and not of a real cause-effect mechanism. This is especially relevant when trying to forecast the future behaviour of the system by means of historical data: relationships between the past and the future are essential, such that the future cannot be forecast from the past if the system changes its structural characteristics (i.e. if there is a non-stationarity in the parameters’ space).
In the talk, the concept of stationarity was reviewed through different simple examples drawn from actual air transport problems. Lastly, a set of possible solutions was discussed, including the use of detrending techniques.
Written by Innaxis Researchers Massimiliano Zanin and Hector Ureta