Computationally Efficient Feature Significance and Importance for Machine Learning Models
This provides a general tool for feature selection and importance assessment in machine learning, though it appears incremental as it builds on forward-selection methods.
The paper tackles the problem of identifying statistically significant features and interactions in machine learning models without requiring model refitting, resulting in a computationally efficient test that applies broadly across model types and tasks.
We develop a simple and computationally efficient significance test for the features of a machine learning model. Our forward-selection approach applies to any model specification, learning task and variable type. The test is non-asymptotic, straightforward to implement, and does not require model refitting. It identifies the statistically significant features as well as feature interactions of any order in a hierarchical manner, and generates a model-free notion of feature importance. Experimental and empirical results illustrate its performance.