How many features does XGBoost work with?


I have been using XGBoost very successful for Finance Fraud detection. That dataset had only 19 features. I wonder how XGBoost will perform with lets say 1000 features or 10,000 features?

Are there any experiences?



I’ve personally tested it with up to 3,5 million features and I’m sure others have used more high-dimensional data.


@thvasilo Did it have good results compared to other methods? Did you test and compare it with fully connected layers?


Algorithm selection is highly problem specific. For a comprehensive survey take a look at this paper.

We won’t be able to help you with that here, we try to focus on XGBoost-specific issues.


Thankf for the answer @thvasilo