AIExplainer
Machine Learning Intermediate 2 min read

What is permutation variable importances?

A measure of the importance of each input variable in a machine learning model

Permutation variable importances is a technique used to evaluate the contribution of each input feature to the predictions made by a machine learning model. It works by randomly permuting the values of one input feature at a time and measuring the increase in model error. The more the model error increases, the more important the feature is.

Think of permutation variable importances like trying to understand how each ingredient in a recipe affects the final dish. If you randomly swap the amount of one ingredient and the dish turns out badly, you know that ingredient is important. If the dish remains the same, the ingredient is less important.

For example, in a model that predicts house prices based on features like number of bedrooms, square footage, and location, permutation variable importances might show that the location is the most important feature, followed by square footage.

Permutation variable importances is used to identify the most relevant input features in a dataset, which can help improve model performance, reduce overfitting, and provide insights into the relationships between variables.

One common misconception is that permutation variable importances measures the correlation between input features and the target variable. However, it actually measures the feature's contribution to the model's predictions, which can be different from correlation.

Permutation variable importances was first introduced in the 2000s as a technique for evaluating feature importance in random forests, but it has since been applied to other machine learning models as well.

feature importance variable importance permutation importance

Three products for different needs — explore what’s relevant to you.