Empirical Risk

Last updated Sep 25, 2022 Edit Source

• Cs ml

When you classify the points in the dataset as $D = {(x_i, \omega_i)}_{i=1}^N$ and you assign each of the points to a class $\hat{\omega_i}$. Then the total empirical risk on the dataset, given the misclassification cost of each class is:

$$ R = \frac{1}{N}\sum_{i=1}^N \lambda_{\omega_i, \hat{\omega_i}} $$