Special thanks to the hosts for providing such a fun challenge.We are grateful for the opportunity to participate and learn.

here are the main ideas for our solution:

-Feature Selection

this was a tough one to be honest, we tried many techniques to select good features but they didn't work. but finaly we decided to drop feature with 85% or more missing values

-Feature Preprocessing

we applied simple label encoding for the columns "city" and "country"

-Feature Generation

we just added time related features since we noticed that they are important for the model

-Key to win

we applied post processing to the predictions by calculating the mean of the target, grouped by the day of the year.

we applied this postprocessing because we noticed that the standard deviation of the target grouped by day of the year is quite low.

here is the link for the our eda notebook.