Hinge Loss - Yousef's Notes
Hinge Loss

Hinge Loss

A measure of the difference between the true labels and the predicted labels in binary classification tasks, commonly used in support vector machines (SVMs). $$ \text{Hinge Loss} = \frac{1}{n} \sum_{i=1}^{n} \max(0, 1 - y_i \hat{y}_i) $$
  • ${y_i}$: true label ({+1} or {-1})
  • ${\hat{y}_i}$: predicted label

#Purpose

Measures the performance of a binary classification model; lower values indicate better performance.

#Properties

  • Non-negative: Hinge Loss is always zero or positive.
  • Encourages a large margin between classes.