Dear Saffas,
The second Absa challenge is officially open. Thank you for your patience while we reviewed and worked with Absa to create a meaningful challenge for them.
Absa has a model they use for this problem but they would like to see if you can beat it and if you come up with interesting features and approaches.
Best of luck!
Thanks amy ... pffft - of course we'll beat it. Can you give it's performance as a benchmark?
Heya @amyflorida626 - well, look at that.
It took @FlatEarthSociety a few hours to build a model that will tell you the answer within a few rands of the truth, and then @Yudheezus improved on that a few hours later. And there is still 3 months to go ... by the end of this we'll have it down to a few cents.
customer.csv contains the income group information for the test sample. did you guys deliberately include this?
Hey @skaak, @cobusburger.
I think Absa has messed up the data again. Think the test set is in thousands while the Train is in nominal (20000 vs 20). Might be wrong but that's how I've got my score.
@cobusburger, Have included it (income gorup) but wish it wasn't present tbh. Just personally think it would have made it a little more interesting but think Absa will want the value of a model that improves their current classification.
Yud!!! Thanks for sharing, this is really helpful for preparing a sub. Could be side-effect because of those commas, I keep on doing something like
Absa included the income group as when you create an account you have to indicate which income group you. However, you don't have to update your income group with salary raises, etc. So I assume most users will stay in their income group however some would be out of them.
Great find @Yudheezus, thank you.
you are correct @amyflorida626 only about 20% of incomes in the test sample are within income grouping range. my bad i thought it would be 100%. appreciate the feedback.
thanks for sharing your insights @Yudheezus, so are theory is that they only read the numerical value up and till the comma delimeter.