Regression loss metrics

You cannot use accuracy for a regression model. Instead, you look at the loss function, the measurement of error.

Selecting the best loss function depends on several factors: * Mean squared error (MSE) Distribution of your target is Gaussian Punishes model for making larger mistakes more than smaller mistakes, but also means it doesn’t handle outliers well Always positive * Mean absolute error (MAE) Target is mostly Gaussian but has some significant outliers ** Handles outliers well, but computationally expensive

Loss functions aren’t the only way to evaluate regression models. R2 or R-squared measures the percentage of data that fits the model. The closer the R-squared value is to 1, the better predictions.

Was this page helpful?