Challenge #380: Inspire 2023 Grand Prix (Round 3)

[isobeltaylor]

complete

[JosephSerpis]

Challenge Completed

Spoiler

[JasperMB]

Spoiler

 

[aatalai]

Nice to have a predictive challenge but would be better if we had to split the dataset, and compare it with other models and choose the best one

[Erin]

Submission attached

Spoiler

For funsies, I played around with oversampling the Unsuccessful deliveries and used a create samples tool. It seemed the default Random Forest without oversampling or splitting the data worked better. I was surprised by that. Guess that's why people create multiple models! 

[lynnesonney]

@AYXAcademy Are you able to provide any insight as to why the goal tool is configured for oversampling? the ACE cancellation file wasn't modified prior to pushing through the forest model tool.

Also - the Pct of total doesn't make sense; I'm showing 3709 / 20093 is 18% not 23%; you get 23 by dividing unsuccessful / successful. Can you help me understand this?! ☺

thanks! 

[JoachimCaronTIL]

Here is my solution

Spoiler

 

[Jean-Balteryx]

Here is my solution !

[JoachimCaronTIL]

Here is another workflow showing comparison between Random Forest with or without oversampling the unsuccessful field value

Spoiler

 

[sfarnham1]

solution