Step by Step
MSE
MSE — punishes large errors much more than small ones
By squaring each error, MSE penalizes large mistakes disproportionately more than small ones, making it the right choice when large errors are especially costly or dangerous.
Example: predicting structural safety margins in engineering, where one large error could be catastrophic — MSE's disproportionate penalty for big mistakes fits this use case well.
MAE
MAE — treats all errors equally, robust to outliers
By using absolute (unsquared) differences, MAE weighs all errors equally regardless of size, making it more robust to outliers than MSE.
Example: predicting typical delivery times where an occasional extreme outlier shouldn't disproportionately dominate the overall error metric.
RMSE
RMSE — original units, still penalizes large errors
The square root of MSE, bringing the metric back into the original units (like MAE) while still retaining MSE's characteristic of penalizing large errors more heavily.
Example: reporting RMSE in dollars for a house price model, giving an interpretable number while still reflecting a stronger penalty for large mispredictions than MAE would.
R²
R-squared — proportion of variance explained
Represents how much of the variance in the target variable the model explains: 1.0 is perfect, 0.0 means no better than predicting the mean, and it can go negative if the model performs worse than that naive baseline.
Example: an R² of 0.85 means the model explains 85% of the variance in house prices, a generally strong result.
Applied Walkthrough
1
A company is building a model to predict structural load tolerances for a bridge design, where a single large prediction error could have catastrophic real-world consequences.
2
Ask: should this problem prioritize MSE or MAE? MSE, since it disproportionately penalizes large errors, aligning with the fact that large mistakes here are especially dangerous.
3
Contrast: if the company were instead predicting typical package delivery times, where occasional outliers shouldn't dominate the metric, MAE would be the more appropriate, more robust choice.
4
In both cases, RMSE could be reported as well, giving an original-units number that still reflects some of MSE's large-error sensitivity, making it a commonly reported middle-ground metric.
Exam Application
Exams test whether you can match the correct regression metric to a described business scenario, specifically based on whether large errors are especially costly (favoring MSE/RMSE) or whether robustness to outliers matters more (favoring MAE).
⚠ Common Trap
The most common trap is assuming one metric is universally "better" than the others. The correct choice depends entirely on the specific business problem — whether large errors deserve disproportionate punishment (MSE/RMSE) or whether all errors should be weighted equally regardless of size (MAE).
✓ Quick Self-Check
1. Why does MSE penalize large errors more than small ones?
Because it squares each error, so larger errors contribute disproportionately more to the total.
Tap to reveal / hide
2. Why is MAE considered more robust to outliers than MSE?
Because it uses absolute (unsquared) differences, weighing all errors equally regardless of size.
Tap to reveal / hide
3. When should you prefer MSE or RMSE over MAE?
When large errors are especially costly or dangerous, such as in structural safety predictions.
Tap to reveal / hide
4. When should you prefer MAE over MSE?
When you want robustness to outliers and don't want occasional extreme errors to dominate the metric.
Tap to reveal / hide
5. What does an R² value of 0.0 represent?
A model that performs no better than simply predicting the mean value for every example.
Tap to reveal / hide