NA values treatment

[PiotMik]

Perform analysis and imputation/removal of non-available values

• [x] ArrivalDelay missingness
• [x] 0s in customer note values

[PiotMik]

NAs are present only in ArrivalDelay variable. Given high collinearity with DepartureDelay this feature will be dropped anyway.

[PiotMik]

• NAs in customer note variables constitute about 8% of data rows.
• Imputation via a multinomial model (to predict artificially a note 1-5) would be too much effort given how small fraction of data that is
• There is also the option to impute missing values by the median passenger score - but I never liked the idea of overrepresenting an artificial "fixed" value in the dataset. You can never be sure the degree of effect that will have on the models.
• Given the small fraction of NAs and around 120,000 non-NA observations left after NA dropping I would recommend just dropping them - we have lots of genuine, high-quality observations to work with.

[adamszczerba]

Check if the data in incomplete records has the same distribution as in the complete ones.

[PiotMik]

Closing, as no negative feedback received on NA treatment.