R², or coefficient of determination, is an [[Error Metrics|error metric]] for [[Regression|regression problems]]. It ranges from 1 (a perfect score) and can be negative (since a model can be arbitrarily worse). In a general case, when the dependent variable is non-constant, a R² measure of 0 means that the model is predicting the average, ignoring the input variables. A very simple way to understand what R² is measuring is the idea of **explained variation**: a R² of 0.8 could be understood as a regression model capturing 80% of the information needed to explain the [[Dependent Variable|dependent variable]], while having 20% of it still missing from it.