Inundata: Mapping Floods in South Africa
'%3e%3cpath fill='%23001489' d='M0 0v6h9V0z'/%3e%3cpath fill='%23e03c31' d='M0 0v3h9V0z'/%3e%3cg stroke='%23fff' stroke-width='2'%3e%3cpath id='d' d='m0 0 4.5 3L0 6m4.5-3H9'/%3e%3cuse xlink:href='%23b' stroke='%23ffb81c' clip-path='url(%23c)'/%3e%3c/g%3e%3cuse xlink:href='%23d' fill='none' stroke='%23007749' stroke-width='1.2'/%3e%3c/g%3e%3c/svg%3e)
Helping South Africa
$10 000 USD
Completed (~1 year ago)
Classification
1340 joined
315 active
Start
Nov 22, 24
Close
Feb 16, 25
Reveal
Feb 17, 25
Are the composite images helpful ???
Help · 17 Dec 2024, 14:50 · 4
Hello guys , hope you are all doing great . I am writing to ask if you find the composite images helpful for improving your model performance ???
For the top scorers @Koleshjr @private_1x @Papito could you please tell us how do you handle the extremely class imbalance .
label 0 491702 1 318 Name: count, dtype: int64
I have tried using SMOTE , undersampling the majority classes but it has significantly increases my Log Loss score from 0.0048 to 0.33 .
My opinions based on what has worked for me
* don't handle the imbalance , leave it as it is and use a good cross validation that takes into account the class imbalance.
* Yes the composite images are important
Thanks for the answers , I truly appreciate it .
I've run the starter notebook twice: once with both the datasets and once with just the rainfall data set and the difference is very slight, not far away from the benchmark. I've run my one notebook, just with rainfall dataset with an hybrid deep learning model achiving a not so bad performance of 0.0056. I've not dealt with imbalance as yet. I'm planning to apply some Feature Engineering
Thank you very much for your insights @dueprincipati. I have also noticed that the images are not very helpful that is why I decided to solely focus on the rainfall data for training my model .
Regarding the class imbalance whenever I tried to handle it my model performance significantly decreases I have end up by leaving the class imbalance as it is and it seems to work .