Feature Importance

Definition

The feature importance of a feature $x_i$ used for Decision Tree is defined as $\mathcal{I}(x_i)=\frac{\sum _{t\in T(x_i)}IG(t)}{\sum _{j=1}^p\sum _{t\in T(x_j)}IG(t)}$ where $IG(t)$ is the Information Gain of a node $t$, and $T(x_i)$ is a set of nodes using the feature $x_i$