Relative Error

INFO

Measures the deviation between predicted and actual physical quantities relative to actual values.

How It Works

$$\text{Relative Error}=\frac{|y-\hat{y}|}{|y|}$$

• $y$: Actual physical value
• $\hat{y}$: Predicted value

This metric expresses the error as a fraction of the true value, making it scale-aware.

What to Look For

• Lower values indicate better physical fidelity
• Useful when absolute error is less meaningful due to varying scales
• Sensitive to small denominators → avoid when $y\approx 0$

Application Models

• pinns
• scientific-ml-models