Mean Squared Error (MSE

Statistical Theory of Regression Loss Functions: MSE, RMSE & MAE

In machine learning and econometric forecasting, evaluating prediction accuracy requires balancing sensitivity to outliers against interpretability:

• Mean Squared Error: Penalizes large errors quadratically (an error of 10 units contributes 100 to the loss, whereas an error of 1 unit contributes only 1). It is differentiable everywhere, making it the primary objective function for gradient-based optimization.

• Root Mean Squared Error (RMSE): Converts MSE back into the original units of the target variable y, representing standard deviation of residuals.

• Mean Absolute Error: Penalizes errors linearly, making it far more robust to extreme anomalies and heavy-tailed data distributions.

• RMSE vs. MAE Ratio: Because RMSE MAE, the gap RMSEMAE serves as a direct indicator of variance in error magnitudes (large values reflect sporadic extreme outlier prediction failures).

How to Use This Calculator

Enter your Actual Ground Truth Values (y - comma or space separated) (e.g. 24, 30, 38, 42, 50, 62, 70) and Predicted Model Values (ŷ - comma or space separated) (e.g. 22, 33, 35, 45, 48, 65, 68). Results update instantly as you type — adjust any value to compare scenarios.

The Formula

Regression Loss & Forecasting Error equations:

Mean Squared Error (MSE): MSE = (1) / (n) Σ_i=1^n (y_i - y_i)^2

Root Mean Squared Error (RMSE): RMSE = √(MSE) = √((1) / (n) Σ_i=1)^n (y_i - y_i)^2

Mean Absolute Error (MAE): MAE = (1) / (n) Σ_i=1^n |y_i - y_i|

Mean Absolute Percentage Error (MAPE): MAPE = (100\%) / (n) Σ_i=1^n | y_i - y_iy_i |

Normalized RMSE (NRMSE): NRMSE = (RMSE) / (y), Mean Error (Bias): ME = (1) / (n) Σ_i=1^n (y_i - y_i)

Worked Example

For example, comparing actual values \ against predictions \ across \ points: Errors are . Squared errors are [4, 9, 9, 9, 4, 9, 4] with sum

• The Mean Squared Error is . The Root Mean Squared Error is . Mean Absolute Error is . With \ and .

Understanding Your Result

Statistical results require context: a mean alone is misleading without knowing the standard deviation (spread). Normal distribution reference: 68% of values fall within ±1 SD of the mean, 95% within ±2 SD, 99.7% within ±3 SD. For small sample sizes (n < 30), use t-distributions rather than z-scores. Statistical significance (p < 0.05) does not imply practical significance — effect size (Cohen's d, R², odds ratio) measures real-world importance. Correlation coefficient r: |r| < 0.3 weak, 0.3–0.7 moderate, > 0.7 strong.