This document describes the ML.FEATURE_IMPORTANCE function, which lets you see the feature importance score. This score indicates how useful or valuable each feature was in the construction of a boosted tree or a random forest model during training. For more information, see the feature_importances property in the XGBoost library.


ML.FEATURE_IMPORTANCE(MODEL `project_id.dataset.model`)


ML.FEATURE_IMPORTANCE takes the following arguments:

  • project_id: Your project ID.
  • dataset: The BigQuery dataset that contains the model.
  • model: The name of the model.


ML.FEATURE_IMPORTANCE returns the following columns:

  • feature: a STRING value that contains the name of the feature column in the input training data.
  • importance_weight: a FLOAT64 value that contains the number of times a feature is used to split the data across all trees.
  • importance_gain: a FLOAT64 value that contains the average gain across all splits the feature is used in.
  • importance_cover: a FLOAT64 value that contains the average coverage across all splits the feature is used in.

If the TRANSFORM clause was used in the CREATE MODEL statement that created the model, ML.FEATURE_IMPORTANCE returns the information of the pre-transform columns from the query_statement clause of the CREATE MODEL statement.


You must have the bigquery.models.create and bigquery.models.getData Identity and Access Management (IAM) permissions in order to run ML.FEATURE_IMPORTANCE.


ML.FEATURE_IMPORTANCE is only supported with boosted tree models and random forest models.


This example retrieves feature importance from mymodel in mydataset. The dataset is in your default project.

  ML.FEATURE_IMPORTANCE(MODEL `mydataset.mymodel`)

What's next