Air Quality Prediction Challenge
Can you use Sentinel 5P data to predict air quality in Kampala for AirQo?
$3 000 USD
Ended 4 months ago
258 active · 663 enrolled
Good for beginners
3rd position
Help · 3 Oct 2022, 15:59 · 10

This competition was our first as a team and a really tough one. Our solution had a few features, I think it wasn't more than 40. We dropped a lot of features which contribute to overfitting especially the angles(solar and sensor), The average pm2.5 reading was a very useful feature from the feature importance plot. We also used the dry&wet season feature. We had a score of 12.20 but unfortunately we didn't select it before the competition ended but the score we chose earned us the 3rd position. Thank you Zindi.

Team Arrow

Discussion 10 answers

Bro that's amazing. 12.20 from just 40 features. Unbelievable. Well done guys well done 👏👏👏

3 Oct 2022, 16:04
Upvotes 0

I will check that again but I know it was less than 50

Nice. And by average pm2_5 you mean mean target encoding?

What cv did you guys use ?

3 Oct 2022, 16:05
Upvotes 0

We just held out the last 20 days when we sorted by date.

Amazing work!

what did you use for feature selection?

3 Oct 2022, 19:16
Upvotes 0

congratulations guys!

8 Oct 2022, 03:04
Upvotes 0

Hey @NIRAN . I have been trying to follow your approach and it's really interesting. You said average pm2_5 really helped you guys. How did you calculate that ? Given that in the test we never had pm2_5?

10 Oct 2022, 15:18
Upvotes 0

By avg pm2.5, I meant rolling mean of 180 days and took lags of