Decision trees can not only be visualized to inspect the decision path for a given feature, but also provide a summary measure of the contribution of each feature to the model fit to the training data. 

The feature importance captures how much the splits produced by the feature helped to optimize the model's metric used to evaluate the split quality, which in our case is the Gini Impurity index. A feature's importance is computed as the (normalized) total reduction of this metric and takes into account the number of samples affected by a split. Hence, features used earlier in the tree where the nodes tend to contain more samples typically are considered of higher importance.

The following chart shows the feature importance for the top 15 features: