Going through the data and reading the description I also think that it is a Multi-Class-Classification-Task.What I find confusing is the Evaluation-part on the Descriptionpage:The rows do not sum up to 1. IMO this would speak against Multi-Class-Classification.
Transaction_ID Bills & Fees Data & WiFi Education Emergency ...
ID_aclde4g8 0.93 0.21 0.78 0.34 ...
ID_gzsdiej7 0.45 0.78 0.12 0.09 ...
I tried a simple submission with all entries==1. Not only did that not throw an error but gave a relatively good score of 2.56. That is kind of confusing if you compare that to the Benchmark == 13.62.
Definitely, a multiclassification challenge IMO
Going through the data and reading the description I also think that it is a Multi-Class-Classification-Task.What I find confusing is the Evaluation-part on the Descriptionpage:The rows do not sum up to 1. IMO this would speak against Multi-Class-Classification.
I tried a simple submission with all entries==1. Not only did that not throw an error but gave a relatively good score of 2.56. That is kind of confusing if you compare that to the Benchmark == 13.62.
Something Serious Needs to be done to the scoring on zindi... It is for most part disappointing
You should submit probabilities, not exact labels. Try this approach and it will increase your score dramatically
How do I get the proba submission
Just train a model with an option of prediction probability.
For instance, you can take a look at the Random Forest approach ( use predict_proba method after training)
I will try using that