Flight Delay Forecast

Goal

Predict aircraft delays in airport terminal area using machine learning techniques on historical data.

Focus

  • On departure and arrival delays of flights for 227 European airports.
  • Airlines: Charter, Low-cost, Traditional Scheduled, All-Cargo, Business Aviation.

Data sources

Traffic, time, and waiting times from 2015 to mid-2018

Weather from 2015 to mid-2018

Departure model & results

Trained on 227 ICAOS
Plotted 4 ICAOs

See F1-score»

Arrival model & results

Trained on 227 ICAOS
Plotted 4 ICAOs

See F1-score»

Additional experiments

In order to make a comparison, it was decided to run the same trainings using the same models and datasets on only 10 ICAOs, the most recurring ones within the entire dataset.

Training on 10 ICAOS: LSZH, LTBA, LEBL, EKCH, ENGM, LOWW, LTFJ, EGKK, LFPO, ESSA

Departure

Trained on 227 ICAOS

Trained on 10 ICAOS

Arrival

Trained on 227 ICAOS

Trained on 10 ICAOS

Conclusions

The model should be refined for being used in a real environment. The performance related to some airports should be improved.

Soul Software model’s results are comparable with the those provided by Eurocontrol in the 2021.

Future developments

The performance should improve through an enrichment of the dataset with weather information, airports features, etc...