R-squared is not commonly used as an evaluation metric for logistic regression models because it works differently for linear and non-linear models. Here are some key points about R-squared in logistic regression:
In linear regression, R-squared represents the proportion of variance in the dependent variable that is explained by the independent variables. It ranges from 0 to 1.
However, for logistic regression the dependent variable is binary/categorical so the concept of variance doesn’t apply directly.
Logistic regression predicts probabilities rather than actual values, so the interpretation of R-squared is different.
Some variations of R-squared have been adapted for logistic regression but they do not have the same probabilistic interpretation as in linear models.
Pseudo R-squared metrics like Cox & Snell, Nagelkerke can range from 0 to 1 but may exceed 1 which is not desirable.
AUC (Area Under the ROC Curve) is a more robust metric to assess logistic regression performance as it doesn’t rely on assumptions behind R-squared.
Other classification metrics like accuracy, precision, recall are also more suitable than R-squared.
So in summary, while R-squared can be calculated for logistic regression, it doesn’t have the same straightforward interpretation and probabilistic meaning as for linear regression models. Alternative metrics are preferred for evaluation.
