Yes, why saying ‘deal with outliers’ if they will not be considered as outliers in the test set?

Thank you for this clarification @Koleshjr. However, this doesn’t change the fact that there was a misleading statement.

Well , in as much as I also had not read the discussion by amy indepth, I don't see any misleading statement since she said, "in the private leaderboard not the public leaderboard " so public lb still had the outliers but since the private lb is what matters, they said they are going to take it into consideration which they have . Well wish I had read that discussion in details , too late now

Yes, you could see it that way. However, for me, the statement ‘It is up to you to determine the best way to deal with outliers in the datasets, both in your modeling and your predictions’ was quite misleading.

Btw great work in detecting the outliers, How did you go about it? I would really love to know, because that was impressive

It was misleading because, for some reason, they didn't want to directly tell us that the data contained errors which would render any model useless.

The way I interpreted this statement was: 'We can't remove outliers from the public dataset; it's too late for that. However, they will be removed from the private test.'

To be honest, the proper way to handle the situation would have been to cancel the competition, fix the data, and start over.

Regardless, the best strategy to win this competition was to probe the entire public test and, for the private test, make predictions as if the outliers didn’t exist.

Thank you @Koleshjr. I will organize my work and share it in another thread!

If they did care about the valuable time of all competitors, they could simply state explicitly in the forum that outliers would be removed from private data.

Your interpretation was correct @yanteixeira , but I believe that in such a competition, clarity is essential to avoid any misunderstanding.

Thank you @kamelyamani , will be waiting for it

@yanteixeira yeah sure , I agree with you. But the problem was in misinterpretation. And after@kamelyamani has clarified how he interpreted is , true that statement could have been interpreted in different ways. A more straightforward answer like: "The private test set does not contain outliers" would have helped and atleast with that the best model could have won ( a fair ground for all). I feel bad for all of us who assumed that the private test also had outliers and decided to post process. I am pretty sure some solutions withput post processing the private test would have ranked way higher.

Yes, the statement should be clear.

I think Zindi and other platforms underestimate the time we spend in competitions. We pour our souls into the problem and suddenly find ourselves affected by a miscommunication issue. It's not fair at all.

This is very accurate and honest position😅

I agree with you. I believe all serious competitors here now know all the techniques to deal with outliers x)