This is the base code use by team zestful-mango. Two approaches were tested:
- Using two gradient boosting models, one for flights with good quality trajectory information, and the other one for the flights with noisy data in the trajectory.
- Using a LSTM neural network with the high quality trajectory data. However, due to lack of time, this approach was not concluded.
For both approaches we develop a common ETL process that stores a datamart (metadata) of the flight pieces and a clean version of the trajectory to use in the LSTM.