Regression models predict numbers. Several metrics measure how close predictions are.
Mean Absolute Error (MAE)
Average absolute difference between prediction and actual value. Easy to interpret in original units, and less sensitive to outliers.
Root Mean Squared Error (RMSE)
Square root of the average squared error. Penalises large errors more heavily. Use when big mistakes are especially costly.
Mean Absolute Percentage Error (MAPE)
Average error as a percentage of actual values. Intuitive for business users, but undefined when actual values are zero and unstable near zero. Alternatives include symmetric MAPE and weighted percentage errors.
R² (Coefficient of Determination)
Proportion of variance explained by the model compared with predicting the mean. 1 is perfect; 0 is no better than the mean; negative is worse. Useful for comparison, but can mislead on its own.
Choosing
- Match the metric to business costs.
- Report more than one metric.
- Compare against a simple baseline, such as the mean or last value.
Look at Residuals
Plot errors against predictions and features to spot patterns that summary metrics hide.