Mean Squared Error (MSE)

#machine-learning #loss-in-ml #interview #evaluation

AKA MSE
AKA L2-loss
MSE is used in Regression problems
Sensitive to large errors for the squaring
Gradient decreases as loss gets closer to 0

Mean Squared Error Formula

$M S E = \frac{1}{n} \sum_{i = 1}^{n} (Y_{i} - {\hat{Y}}_{i})^{2}$

Pros:

Convex function, so one Global Minima
No Local Minima
If error is big, penalizes higher by squaring them
Good for predicting outliers as give more weight to them

Cons:

Bad at Handling Outliers as it will give more weights to the outliers
1. better to use Mean Absolute Error (MAE)
Same error for one big error and many small errors

Related Notes