Mean squared error
Also: MSE, L2 loss, squared error
MSE = (1/n)·Σ(yᵢ − ŷᵢ)². Squaring makes all errors positive and penalizes large errors much more than small ones, which makes MSE smooth to optimize but sensitive to outliers. Its square root (RMSE) is in the target’s units.
MSE = (1/n) Σ (yᵢ − ŷᵢ)²
Example
Predictions (3, 5) against targets (2, 7) have errors 1 and −2, so MSE = (1² + 2²) / 2 = 2.5 and RMSE ≈ 1.58.