Course contentsShow
← Machine Learning and Deep Learning
Lesson 470 of 3,53811. Model Evaluation MetricsFree lesson

Mean Squared Error (MSE) and RMSE

Learn MSE's quadratic penalty for errors, why it penalizes large errors more, and RMSE for interpretability in original units.

Mean Squared Error (MSE) and RMSE

What You'll Learn

You'll learn how MSE penalizes prediction errors quadratically, why it hits large mistakes harder than small ones, and how RMSE converts the result back to original units for easier interpretation.

The Core Idea

Mean Squared Error (MSE) calculates the average of squared differences between your predictions and actual values. Instead of just measuring how far off you are (like MAE), MSE squares each error before averaging.

Why Squaring Matters

Squaring creates a quadratic penalty that makes large errors much more costly:

  • A prediction off by 2 units contributes 4 to MSE
  • A prediction off by 10 units contributes 100 to MSE (25× worse!)

Think of it like falling: stumbling slightly might be fine, but a big fall hurts disproportionately more. MSE treats prediction errors the same way—it's very sensitive to outliers and big mistakes.

From MSE to RMSE

Root Mean Squared Error (RMSE) is simply the square root of MSE. Why take the square root? To return to the original units of your target variable.

If you're predicting house prices in dollars, MSE is in "squared dollars" (not very intuitive!). RMSE converts back to dollars, making it directly comparable to your data and easier to explain to stakeholders.

When to Use MSE vs RMSE

  • MSE: When you want to heavily penalize large errors during optimization (common in loss functions)
  • RMSE: When you need to interpret and communicate error magnitude in familiar units

Both share the same sensitivity to outliers—a single terrible prediction can inflate the metric significantly.

Key Takeaway: MSE squares errors to penalize large mistakes quadratically, making it sensitive to outliers; RMSE takes the square root to restore interpretability in the original measurement units.