Is XGBoost cure to very skewed dataset?

Course Q&A Machine Learning Specialization Advanced Learning Algorithms
,
1

Hi,

I am in XGBoost lecture, and it seems that the method is super good and simple since only resampling is done much smarter.

My question is that when you have a very skewed results, ex: %99.1 negative, %0.9 positive, does this algorithm helps automatically? I

example treatment by sklearn for other methods, I guess that, to estimate class weights for unbalanced datasets:
compute_class_weight

sklearn.utils.class_weight.compute_class_weight(class_weight, ***, classes, y, sample_weight=None )…

Thanks,

2

I am not sure about the internal implementation of the library regarding the imbalanced datasets, but normally, when you are dealing with imbalanced datasets, the rule is to use not just an accuracy indicator but also precision, recall, and F1 score. If you can find the implementation of the library, you can check the internals. If the only metric given as output is accuracy, perhaps it’s not that good, though.

1 Like