XGBoost
XGBoost is a software implementation of gradient boosting for estimating decision trees.
Contents
Usage
XGBoost is written in C++, but generally is used through the official bindings for either R or Python.
Interpretation
Features of an XGBoost model are evaluated according to three attributes:
Gain: relative contribution of a feature to the overall model
Cover: proportion of observations related to a feature
- the number of observations where this feature is the deciding tree split, then normalized across all features
Frequence: proportion of tree splits formed by a feature
All of these attributes are relative, sum to 1, and can be directly compared.