Glossary term
Glossary term
Evaluation and Benchmarks
A regression metric indicating how much variation in a label is due to an individual feature or to a feature set. R-squared is a value between 0 and 1, which you can interpret as follows:
An R-squared of 0 means that none of a label's variation is due to the feature set.
An R-squared of 1 means that all of a label's variation is due to the feature set.
An R-squared between 0 and 1 indicates the extent to which the label's variation can be predicted from a particular feature or the feature set. For example, an R-squared of 0.10 means that 10 percent of the variance in the label is due to the feature set, an R-squared of 0.20 means that 20 percent is due to the feature set, and so on.
R-squared is the square of the Pearson correlation coefficient between the values that a model predicted and ground truth.
Created for this library
A pricing team reports R-squared alongside RMSE to communicate the share of variance the model explains to business reviewers.
An economics research team reports R-squared in its regression analysis of policy effects across markets.
A finance analytics team reports R-squared as one of several diagnostics for its forecasting model.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License