TODO collection for future reference

[wurzelDeveloper]

TODOs:

Add features to improve the prediction. All multiplications and divisions should be features.

Feature ideas:r

• zero fuel weight
• estimated engine type by knowing (Austrian, Swiss, Vueling)
• take off speed
• take off climb rate
• taxi out fuel consumption
• landing speed
• weather at destination
• weather at departure airport
• estimated fuel burn for route (need to account for flight phases)
• airline specific differences => cluster similar data per airline
• route schedule (daily, weekly etc.)
• month of year?

further improvements

• feature importance
• normalize features
• bayesian hyperparameter search
• use neural network as predictor (learn JAX?)
• incorporate uncertainty?!

k-fold cross validation to get a good signal without waiting for the leaderboard

clean data remove quasi duplicates with same tow

https://www.transtats.bts.gov/AverageFare/ https://www.transtats.bts.gov/Data_Elements.aspx?Data=4 https://data.europa.eu/data/datasets/43c6ugqwp92dx7vlgnzja?locale=en Diversion Airports: https://www.bts.gov/topics/airlines-and-airports/domestic-flights-tarmac-times-more-3-hours-and-international-flights-9 Taxi out time: https://www.transtats.bts.gov/ONTIME/OriginDestination.aspx https://www.transtats.bts.gov/ONTIME/Departures.aspx

https://www.transtats.bts.gov/DL_SelectFields.aspx?gnoyr_VQ=FGJ&QO_fu146_anzr=b0-gvzr https://www.bts.gov/browse-statistical-products-and-data/bts-publications/data-bank-28ds-t-100-domestic-segment-data

How to split into regional, buisness, cargo,... https://www.eurocontrol.int/sites/default/files/2022-05/eurocontrol-market-segment-update-2022-05.pdf

https://www.easa.europa.eu/eco/eaer/appendix

fuel flow per engine etc. https://www.easa.europa.eu/en/domains/environment/icao-aircraft-engine-emissions-databank

fuel prices 2022 https://www.kaggle.com/datasets/zusmani/petrolgas-prices-worldwide/data CC0 license - obtained with google

https://data.transportation.gov/Aviation/Consumer-Airfare-Report-Table-1a-All-U-S-Airport-P/tfrh-tu9e/about_data

https://destinationinsights.withgoogle.com/intl/en_ALL/

Likely not allowed to use, but could be checked if results improve dramatically => warrants manual collection of more data https://www.kaggle.com/datasets/heitornunes/aircraft-performance-dataset-aircraft-bluebook/data

Idea from Lukas: Add tax dataset?

Idea for a general plan:

1. We go very broad and extensively leverage open datasets
2. We train a small model to find out which features are most important
3. For these features, we do feature engineerin or include additional datasets from the domain
4. We clean the existing data and features to reduce noise
5. We scale up the model to the biggest possible size

---