I was running your code in my system
Came across this minor issue.

Need to replace

train_df[date_col + "_week"] = train_df[date_col].dt.**year**
test_df[date_col + "_week"] = test_df[date_col].dt.**year**

with

train_df[date_col + "_week"] = train_df[date_col].dt.**week**
test_df[date_col + "_week"] = test_df[date_col].dt.**week**

Hi Sir!

Newbie question.

I see that you have done feature engineering in Club Mahindra competition, but I didn't came across any EDA part. I wanted to know how you came to know these features are important and will work. Is it simple hit and trial and compare performance of model or is there any other way.

Regards
Bhaskar

Variables such as month, Weekday, Hour are all categorical variables. Why haven't you one-hot-encoded them before feeding into the xgboost classifier?