hq38 Final Peer Review

[joyyyq]

Very interesting topic and good report.
Three things I like:

1. The team achieves a nice accuracy
2. The team carefully evaluate their results by both accuracy and recall. They also further analyze why the recall is high (the weather forecast is daily instead of hourly so that a whole day’s flights are likely to be predicted as delayed though only flights in a certain period are likely to be).
3. The team is aware of the drawbacks of different methods. For instance, they mention that both random forest and XGBoost suffer from interpretability and they make adjustments accordingly (for instance by choosing a high number of trees and low maximum search depth).

Three potential improvements:

1. Maybe can try more models and see if the recall can be improved.
2. Like the team mentioned, incorporate hourly weather data and other data other than weather to reduce bias.
3. Provide both qualitative and quantitative prediction about flight delay (provide both Yes/No prediction and the expected amount of delay).